Hot spots of vegetation-climate feedbacks under future greenhouse forcing in Europe

Authors


Abstract

[1] We performed simulations of future biophysical vegetation-climate feedbacks with a regional Earth System Model, RCA-GUESS, interactively coupling a regional climate model and a process-based model of vegetation dynamics and biogeochemistry. Simulated variations in leaf area index and in the relative coverage of evergreen forest, deciduous forest, and open land vegetation in response to simulated climate influence atmospheric state via variations in albedo, surface roughness, and the partitioning of the land-atmosphere heat flux into latent and sensible components. The model was applied on a ∼50 × 50 km grid over Europe under a future climate scenario. Three potential “hot spots” of vegetation-climate feedbacks could be identified. In the Scandinavian Mountains, reduced albedo resulting from the snow-masking effect of forest expansion enhanced the winter warming trend. In central Europe, the stimulation of photosynthesis and plant growth by “CO2 fertilization” mitigated warming, through a negative evapotranspiration feedback associated with increased vegetation cover and leaf area index. In southern Europe, increased summer dryness restricted plant growth and survival, causing a positive warming feedback through reduced evapotranspiration. Our results suggest that vegetation-climate feedbacks over the European study area will be rather modest compared to the radiative forcing of increased global CO2 concentrations but may modify warming projections locally, regionally, and seasonally, compared with results from traditional “off-line” regional climate models lacking a representation of the relevant feedback mechanisms.

1. Introduction

[2] Processes and mechanisms coupling climate and terrestrial ecosystems play an important role in climate change, since interacting feedbacks are likely to modify the magnitude of change in climate. Dynamic vegetation models (DVMs), first developed in the 1990s, enabled structural vegetation changes in response to climate and other key driving forces such as atmospheric CO2 concentrations to be simulated with reasonable accuracy at large (regional-global) scales [Cramer et al., 2001]. Such vegetation changes have been demonstrated to cause potentially important feedbacks on climate by affecting the effectuality of the terrestrial biosphere as a sink for CO2 from the atmosphere [Cox et al., 2000; Friedlingstein et al., 2006] and through changes in a number of feedback mechanisms collectively referred to as biophysical feedbacks, involving plant-mediated changes in albedo, evapotranspiration, surface roughness and the partitioning of the land-atmosphere heat flux into sensible and latent components [Betts et al., 1997; Betts, 2000; Sitch et al., 2005]. This has lead to an increasing awareness of the importance of recognizing the terrestrial biosphere as part of the climate system, and of incorporating the relevant feedback mechanisms in climate models [Denman et al., 2007]. Since CO2 is well mixed in the atmosphere, carbon cycle feedbacks are closely linked to global climate change [Bala et al., 2007] and are often addressed with global Earth System Models (ESMs), that is, General Circulation Models (GCMs) incorporating a DVM submodel. Biophysical feedbacks are to a larger extent controlled by mechanisms of regional character and the relative importance of individual feedback mechanisms may differ among regions [Sitch et al., 2005; Bala et al., 2007]. Climate change studies using regional climate models (RCM) may however miss significant regional biophysical feedbacks that could modify the projected climate, since RCMs do not include vegetation dynamics [Hibbard et al., 2007].

[3] Future scenarios point toward a northward expansion of boreal forests into extant tundra areas [Levis et al., 1999], implying a major long-term structural change in the vegetation cover thought to amplify the warming trend over the affected regions through a reduced albedo especially in the winter months when the dark canopies of the invading trees mask the bright underlying snow. The reduced albedo results in a greater proportional absorption of the incoming solar radiation, causing an increase in reemitted longwave radiation, and warming of the atmosphere above. At lower latitudes, higher temperatures increase the transpirative demand of the atmosphere for water, depleting soil moisture and reducing plant productivity. If warmer temperatures and a changed rainfall regime increase the frequency or severity of drought, reduced survivorship may result. These factors may cause a reduction in vegetation cover, associated with a decrease in leaf area index (LAI, i.e., the ratio of one-sided leaf area to the area of ground they cover), which in turn may amplify the warming through a reduction in the surface area over which plants emit water to the atmosphere in the process of (evapo-)transpiration [Betts et al., 2004]. LAI adjustments also influence the interception of rainfall by vegetation canopies, another component of evapotranspiration. Reduced evapotranspiration equates to a reduction in the proportion of longwave energy returned from the land surface to the atmosphere as latent heat, making more energy available as sensible heat, directly heating the lower atmosphere.

[4] Severe weather episodes in recent years in terms of increasing frequencies in flood and drought events in Europe have been attributed to both increased climate variability and land use changes. The increase in climate variability has so far been presumed to be a consequence of changes in atmospheric circulation as a response to global warming [Meehl and Tebaldi, 2004; Ogi et al., 2005]. Seneviratne et al. [2006], however, found summer climate variability in the central and eastern parts of Europe to be mainly a consequence of strong atmosphere-land surface interactions in terms of soil moisture-temperature feedbacks, suggesting that climate variability due to these feedbacks is likely to increase under future warming as a consequence of the northerly shift of vegetation zones in response to isotherm migration. These findings underscore the role of the land surface in climate dynamics and suggest that climate change in Europe can only be satisfactory simulated using models that include relevant land surface processes and feedbacks.

[5] Here we explore the potential role of vegetation dynamics and biophysical feedbacks in regional climate change by applying a regional climate model framework that incorporates a process-based model of vegetation dynamics, accounting for feedbacks of climate-driven changes in vegetation patterns and LAI on the atmosphere, via changes in albedo and evapotranspiration. On the basis of an AOGCM-forced 21st century simulation over Europe and North Africa, we identify and discuss a number of geographically coherent “hot spots” of vegetation-climate feedbacks, here defined as an area characterized by a fairly homogenous climate and syndrome of biophysical feedback phenomena. Neither uncoupled (offline) models with prescribed land surface data sets, nor global ESMs can be expected to capture such features.

2. Material and Methods

2.1. RCA-GUESS: A Coupled Regional Climate-Vegetation Model

[6] The regional ESM RCA-GUESS constitutes the two submodels RCA3 [Samuelsson et al., 2006], representing the climate part of the model, and LPJ-GUESS [Smith et al., 2001; Wramneby et al., 2008], which provides RCA3 with dynamic updates of vegetation fractions and LAI.

[7] RCA3 (hereinafter “RCA”) is the latest version of the Rossby Centre regional atmospheric model [Kjellström et al., 2005]. The model incorporates an improved land surface scheme (LSS) relative to previous versions of the model [e.g., Jones et al., 2004]. The current LSS [Samuelsson et al., 2006] divides the land surface of each simulated grid cell into an open land and forest tile, the latter in turn divided into a needle-leaved (conifer) and a broad-leaved fraction. The open land tile is assumed to be occupied by herbaceous vegetation only. Bare ground may also occur in either tile. Vegetated fractions with and without snow-lie are further distinguished (Figure 1).

Figure 1.

Weighted winter (December, January, and February) temperature 2m-T (°C) for (a) 1961–1990, (b) change by 2071–2100 in the feedback experiment (“var”), and (c) feedback contribution ΔTvar − ΔTstat, where positive numbers indicate an amplification of future warming with vegetation dynamics and negative values indicate a dampening of future warming.

[8] LPJ-GUESS is a process-based DVM optimized for regional applications that explicitly accounts for heterogeneity in age, size and vigor among cooccurring individuals (trees and a herbaceous understory) and among patches at different stages of development following disturbances in forest landscapes [Smith et al., 2001; Hickler et al., 2004; Wramneby et al., 2008]. In the coupled model, RCA-GUESS, vegetation dynamics and leaf phenology, simulated by LPJ-GUESS, influence climate by affecting the relative cover of different vegetated fractions (needle-leaved forest, broad-leaved forest and herbaceous open land vegetation) and variation in their respective LAIs on seasonal (phenology) and interannual time scales.

[9] The relative cover of different vegetation types affects surface albedo, which is a weighted average of prescribed albedo constants for trees, open land vegetation, snow and other nonvegetated surfaces. For forested areas, the relative contributions of canopy, soil surface and snow are scaled by projective canopy cover, estimated as a function of LAI using Beer's Law. The albedo inversely governs the net uptake of incoming solar radiation by, and temperature of, the land surface.

[10] The return longwave energy fluxes from the land surface to the atmosphere are influenced by the relative cover of the open land and forest tiles in each grid cell, their LAIs, and the influence of these factors on surface roughness and aggregate stomatal conductance. The latent heat flux (W m−2), which corresponds to the energy absorbed and transferred from the land surface to the atmosphere in the process of evapotranspiration, is computed on the basis of the following general equation, a specific variant of which is applied to each cover fraction:

equation image

where ρ is air density (kg m−3), Le is the latent heat of vaporization of water (J kg−1), qs is the surface saturated specific humidity at a given surface temperature Ts, qam is the specific humidity of the lowest layer of the atmosphere (at height 90 m in this study), ra is aerodynamic resistance to water vapor conductance (sm−1) in the boundary layer adjacent to the vegetation or ground surface, and rs is surface resistance to conductance of water vapor (sm−1). In the case of vegetated areas (forests and vegetated open land), the latter term is the inverse of aggregate stomatal conductance, and consequently scales with vegetation LAI:

equation image

where rs,min is a prescribed minimum surface resistance that differs for forest and open land vegetation and the Fj are scalars representing the influence of incoming photosynthetically active radiation (F1), soil water stress (F2), vapor pressure deficit (F3), air temperature (F4), and soil temperature on surface resistance (F5) (full details are given by Samuelsson et al. [2006]). Aerodynamic resistance terms, which are distinguished for within-canopy and boundary layer air, are affected by surface roughness; the relationship is parameterized as a function of LAI, with the consequence that the aerodynamic resistance related to canopy-air boundary decreases with increasing LAI, whereas the within-canopy aerodynamic resistance decreases with increasing LAI.

[11] Sensible heat fluxes are computed for each cover fraction following

equation image

where cp is the specific heat capacity of the air (J kg−1 K−1) and TsTam is the temperature differential between the surface and the lowest layer of the atmosphere. LAI influences the aerodynamic resistance, ra, denser vegetation promoting turbulent mixing and a more rapid transport of heat from the surface to the atmosphere.

[12] A detailed description of LPJ-GUESS is given by Smith et al. [2001]. Formulations of plant physiology, canopy-boundary layer biophysics and ecosystem biogeochemistry are in common with the global DVM LPJ [Sitch et al., 2003]. The version used in this study includes improved formulations of ecosystem hydrology as described by Gerten et al. [2004]. LPJ-GUESS simulates growth and competition among individual trees and a herbaceous understory cooccurring and interacting with one another in competition for space, light and soil resources (water and, implicitly, nutrients). Photosynthesis, respiration, tissue turnover and carbon allocation to leaves, fine roots and stems are modeled on an individual basis. Height and diameter growth are regulated by carbon allocation, conversion of sapwood to heartwood, and a set of prescribed allometric relationships for each represented plant functional type (PFT). Population dynamics (establishment and mortality) are influenced by current resource status, demography, and the life history characteristics of each PFT [Hickler et al., 2004; Wramneby et al., 2008]. Exogenous biomass-destroying disturbances (corresponding for example to storms, fires or forest harvest) are represented as a stochastic process, here with an expectation of 0.01 year−1, corresponding to a local expected return interval of 100 years. The disturbance interval was chosen to reflect reconstructed disturbance histories for Europe [e.g., Zackrisson, 1977; Nagel et al., 2007] and is consistent with previous studies using LPJ-GUESS and other vegetation models [e.g., Prentice et al., 1991; Morales et al., 2007]. In the “cohort mode” employed in the present study, individual trees are distinguished, but are identical within each cohort (age class). As population processes and disturbances are modeled stochastically, stand characteristics are averaged over a number of patches, here 30 per vegetated tile and 0.1 ha in size, representing “random samples” of the simulated stand.

[13] In the coupled model, LPJ-GUESS was called at the end of each simulation day to update the state of each vegetated tile within a grid cell on the basis of the current day averages for the following forcing variables, computed by RCA: air temperature (midcanopy for forest, 2 m above surface for open land), soil temperature at 0.14 m depth, net downward shortwave radiation and soil water content (fraction of field capacity) for the upper (0–0.5 m) and lower (0.5–1.5 m) soil layer distinguished by LPJ-GUESS. The soil hydrology submodule that forms part of the standard, offline realization of LPJ-GUESS was disabled in RCA-GUESS. As RCA employs a more finely divided soil layer profile for hydrology calculations, soil water contents for the two layers used by LPJ-GUESS were estimated as depth-weighted averages of the corresponding layers in RCA. The atmospheric carbon dioxide (CO2) concentration, required for the computation of photosynthesis and stomatal conductance in LPJ-GUESS, was prescribed from an external database (see section 2.2).

[14] Six PFTs were distinguished, representing the major groups of plants occurring naturally and in managed forests across the European and North African study area. The PFTs and the main parameters governing their behavior in the model are summarized in Table 1. The parameter set distinguishing PFTs is based on the work of Morales et al. [2007] with the modifications described by Smith et al. [2010]. For the open land tile, trees were excluded, only C3 herbaceous plants (representing grassland and the crops most commonly planted in Europe) being allowed to grow. State variables for the needle-leaved (NE, MNE) and broad-leaved (TBS, IBS, BE) PFTs were aggregated to provide a single summed value for each of cover class in RCA. Fractional cover of needle-leaved and broad-leaved forest within the forest tile was estimated at the foliar projective cover (FPC) using Beer's law:

equation image

where LAIneedle and LAIbroad are the summed leaf area index of needle-leaved and broad-leaved trees within the forest tile, respectively. The vegetated fraction of the open land tile was estimated in a similar way:

equation image

where LAIgrass is the leaf area index of the simulated C3 grass (typically representing an agricultural crop such as wheat) in the open land tile.

Table 1. European Plant Functional Types and Their Characteristics as They Are Represented in LPJ-GUESSa
 NETBSIBSBEMNEG
  • a

    NE, boreal/temperate needle-leaved evergreen; TBS, temperate shade-tolerant broad-leaved summer green; IBS, boreal/temperate shade-intolerant broad-leaved summer green; BE, temperate shade-tolerant broad-leaved evergreen; MNE, Mediterranean shade-intolerant needle-leaved evergreen; G, grass.

Leaf phenologyevergreenwinter deciduouswinter deciduousevergreenevergreenwinter/drought deciduous
Shade tolerancehighhighlowhighlowlow
Fire tolerancelowlowlowlowlowhigh
Minimum coldest month T for survival (°C)-−18-1.71.7-

[15] The relative coverage of the forest and land tiles (including vegetation-free fractions) was prescribed from an external land cover database, representing present-day vegetation cover due both to natural (biogeographic) factors and human land use (see section 2.2). By default, areas classed as natural vegetation in the database were credited to the forest tile. However, the tile sizes were permitted to adjust dynamically from their prescribed values in the event that the simulated maximum growing season LAI summed across tree PFTs in the forest tile fell below 1, signifying marginal or stunted woody plant growth, for example, due to constrainment of the annual growth period by extreme cold (alpine areas) or drought (dry climate ecosystems of the Mediterranean and North Africa). In this case, the LAI of the simulated trees (taken to represent shrubland- or tundra-like vegetation) was transferred to the open land tile whose vegetation cover was recomputed using equation (5), while the forested fraction was reset to zero.

[16] The extreme reflective characteristics of snow make it a significant interactive component in the vegetation-climate system. In the coupled model, forest canopies are assumed to mask any underlying snow, whereas on open land, the converse applies: snow masks any underlying vegetation. In the forest tile, simulated changes in tree FPC, linked to LAI by equation (4), influence the fractional area of the snow tile and the corresponding snow-masking effect in vegetated areas.

[17] RCA-GUESS was applied at a horizontal resolution of ∼50 km, 24 levels in the atmosphere and a time step of 30 min for physical processes, one day for leaf phenology, and one year for vegetation dynamics (variations in vegetation cover fractions and LAI changes caused by establishment, mortality and individual growth).

2.2. Coupling and Experiment

[18] RCA-GUESS was set up on a rotated latitude-longitude grid at a spatial resolution of 50 × 50 km over Europe. Boundary conditions were taken from a global climate simulation from 1961 to 2100 with the ECHAM5 AOGCM [Roeckner et al., 2006] forced by greenhouse gas emissions from the A1B scenario of the IPCC [Nakicenovic et al., 2000]. The ECHAM5 model was chosen since it has been proven to be qualitatively robust in comparison with many other GCMs in the Fourth Assessment Report (AR4) of IPCC [Van Ulden and Van Oldenborgh, 2006]. The A1B scenario represents anthropogenic emissions consistent with assumptions of globalization and strong economic growth. ECHAM5-A1B has been targeted in the ENSEMBLES project by a number of different RCMs, enabling comparison between different model simulations (www.ensembles-eu.org, www.metoffice.gov.uk).

[19] RCA-GUESS was initialized (“spun up”) to achieve an approximate steady state between the vegetation structure and PFT composition and the initial simulated climate. The climate submodel required a spin-up phase of only a few months, because of the short climate system memory. For the vegetation submodel, a two-stage spin-up was performed. The first stage encompassed 360 simulation years in off-line mode, utilizing monthly values of temperature, precipitation and cloud cover fraction for the nearest 0.5 × 0.5° grid cell from the CRU05 global historical climate database [New et al., 1999]. As the CRU data set only extends back in time to 1901, data for the first 30 years (1901–1930) were cycled repeatedly for the first 300 years of the spin-up. Interannual trends in the mean temperature for a given month were removed using linear regression. The 1901 atmospheric CO2 concentration was assumed for the first 300 years. CO2 concentrations for the period from 1901 until the first year of the coupled climate simulation were taken from the database compiled by McGuire et al. [2001]. From 1961 the simulation continued for a further 30 years in coupled mode, vegetation dynamics influencing the simulated climate and vice versa.

[20] The second spin-up stage was required to ensure a smooth transition from CRU- to RCA-forced vegetation dynamics, as a step change in the forcing climate could lead to artificial vegetation dieback or other sudden changes in structure requiring years or even decades for a return to steady state. A new simulation was performed in off-line mode forced by the time series of output data generated by RCA during the 1961–1990 coupled phase of the initial simulation, cycled repeatedly over 360 years. This was followed by the full coupled simulation, commencing in 1961.

[21] The PFT composition and biomass pools of the simulated vegetation at the end of the spin up is in dynamic equilibrium with the initial RCA-generated climate. Key aspects of the simulated vegetation structure and function (winter and summer LAI, net primary production (NPP), and PFT geographic distributions) have been evaluated in comparison to observations in a separate study [Smith et al., 2010]. They are broadly consistent with results from available validation studies with the offline version of LPJ-GUESS [Smith et al., 2001, 2008; Zaehle et al., 2007; Wramneby et al., 2008].

[22] Two separate simulations were performed, with and without feedbacks of vegetation dynamics enabled. In the first (“feedback”) simulation, the physical and vegetation submodels were coupled throughout the simulation period 1961–2100. The second (“nonfeedback”) simulation began from the state of the feedback simulation in 1991 but differed in that the forcing fields generated by the vegetation submodel were replaced by prescribed daily averages for 1961–1990 from the initial (feedback) simulation. In other words, simulated changes in vegetation structure, composition and cover after 1990 did not affect the simulated climate in the nonfeedback simulation. By comparing the output from the two runs, it was possible to infer the feedbacks of vegetation dynamics on the simulated climate (see section 2.3).

[23] The relative coverage of the forest and land tiles (including vegetation-free fractions) was prescribed from the ECOCLIMAP land cover data set [Masson et al., 2003], representing present-day land cover. The vegetation within each tile was simulated by the vegetation submodel, as described in section 2.1. Changes in land cover due to human land use decisions were assumed not to occur during the simulation.

2.3. Method of Analysis

[24] Hot spots of vegetation-climate feedback were identified on the basis of mapping the anomalies of a number of physical quantities among others; mean surface temperature (T), sensible (H) and latent heat fluxes (E), the Bowen ratio (defined as the ratio of sensible to latent heat land-atmosphere flux, H/E) and albedo (α). The Bowen ratio is a physical quantity strongly dependent on vegetation control of the proportion of the absorbed surface energy that is returned to the atmosphere as the latent energy of water vapor. The underlying process (evapotranspiration) is controlled at the leaf scale by stomatal conductance and at the landscape scale by changes in the distribution and structure (density, leaf area) of different vegetation elements. Hot spots were identified qualitatively and subjectively on maps of the differences between the feedback and nonfeedback simulations in the anomalies of the average value for the last 30 years of the scenario simulation (2071–2100) relative to the control period 1961–1990:

equation image

where Fb is feedback contribution, X is physical variable examined, var is value from feedback simulation (see section 2.2), stat is value from nonfeedback simulation, scen is scenario period 2071–2100, and ctrl is control period 1961–1990.

3. Results

[25] Three latitudinal bands of CO2-induced surface-temperature increases can be distinguished from the results. These are evident in both the feedback and nonfeedback simulations. In winter the largest temperature increase by 2071–2100 takes place in northern Europe (3–7°C, with the largest increase in the very north) (Figure 1b), whereas Mediterranean areas experience the largest temperature increase in summer (3–7°C, with the largest increase in the very south) (Figure 2b). The temperature changes are associated with a response pattern in vegetation dynamics, which in turn feeds back to climate, amplifying or dampening the temperature increase in different parts of the study domain (Figures 1c and 2c). The strength of the investigated feedbacks varies with season (Figures 1c and 2c).

Figure 2.

Weighted summer (June, July, and August) temperature 2m-T (°C) for (a) 1961–1990, (b) change by 2071–2100 in the feedback experiment (“var”), and (c) feedback contribution ΔTvar − ΔTstat, where positive numbers indicate an amplification of future warming with vegetation dynamics and negative values indicate a dampening of future warming.

3.1. Vegetation Response to Simulated Climate Change

[26] In general, climate warming and CO2 fertilization [Hickler et al., 2008] lead to a positive response of, particularly, woody PFTs. A general northeasterly shift of vegetation zones is evident (Figure 3). Boreal needle-leaved evergreen (conifer) forest replaces tundra and expands into mountain areas of Fennoscandia (Figure 4). Temperate deciduous broad-leaved trees expand into regions previously entirely occupied by boreal conifer forests. Mediterranean and temperate drought-tolerant evergreen vegetation expands into areas currently dominated by deciduous forest. A minor increase in the abundance of drought-tolerant woody PFTs is simulated in forested areas of the Mediterranean. Open-land areas in the same region, however, exhibit reductions in vegetation cover and leaf density (Figure 4).

Figure 3.

Normalized phenology index Cp = (LAIeg − LAId)/(LAIeg + LAId) [Wramneby et al., 2008] quantifying the relative abundance of evergreen (eg) and deciduous (d) PFTs on the basis of annual LAI in (a) 1961–1990 and (b) 2071–2100. Increasing positive values equal increasing eg dominance, negative values equal d dominance, and values around 0 imply an even mixture. (c) The percentage of change in Cp by 2071–2100.

Figure 4.

Normalized physiognomy index Cp = (LAIw − LAIg)/(LAIw + LAIg) [Wramneby et al., 2008] quantifying the relative abundance of woody (w) and grass (g) PFTs on the basis of annual LAI in (a) 1961–1990 and (b) 2071–2100. Increasing positive values equal increasing w dominance, negative values equal g dominance, and values around 0 imply an even mixture. (c) The percentage of change in Cp by 2071–2100.

3.2. Vegetation Feedbacks on Simulated Climate

[27] Feedbacks caused by shifts in the distributions of different vegetation types are evident in at least three geographically coherent parts of the European study area. As the analysis below indicates, the main feedback mechanisms involved are somewhat distinct for each of these areas, fulfilling our definition of a “feedback hot spot” (see section 1). The hot spots are (1) the Scandes mountain range in Scandinavia, where albedo reduction associated with tree-line shifts amplifies warming (Figure 5); (2) central and eastern Europe, where enhanced vegetation production and leaf area promotes an increased partitioning of the land-atmosphere energy exchange to latent heat, mitigating near-surface warming (Figure 6); and (3) southern Europe, where drought stress in summer leads to sparser vegetation and leaf shedding in summer that amplifies the climate-driven reduction in evapotranspiration, exacerbating near-surface warming (Figure 6).

Figure 5.

Weighted winter (December, January, and February) albedo for (a) 1961–1990, (b) change by 2071–2100 in the feedback experiment, and (c) feedback contribution Δalbedovar − Δalbedostat, where positive numbers indicate a dampening of future warming with vegetation dynamics and negative values indicate an amplification of future warming.

Figure 6.

Weighted summer (June, July, and August) latent heat flux E (W m−2) for (a) 1961–1990, (b) change by 2071–2100 in the feedback experiment, and (c) feedback contribution ΔEvar − ΔEstat, where positive numbers indicate a dampening of future warming with vegetation dynamics and negative values indicate an amplification of future warming.

[28] In the Scandes, CO2-induced warming results in an expansion of forested areas onto the high mountains where low-growing tundra vegetation (represented by the herbaceous PFT “Grass” in the model) is simulated under the present-day climate. The first trees to appear are the shade-intolerant broad-leaved summer-green PFT, corresponding to the tree line forming mountain birch (Betula pubescens ssp. tortuosa). These are followed by boreal needle-leaved evergreen (NE), corresponding to the European Boreal forest dominants Scots pine (Pinus sylvestris) and Norway spruce (Picea abies), which dominates the simulated mountain forest by the end of the 21st century (Figure 7). The tree-line advance causes an additional reduction in the albedo (in addition to climate driven snowmelt) in winter (Figure 5), indicating that vegetation dynamics amplify climate driven warming in this region (Figure 1c). The results suggest an additional temperature increase in winter (DJF) between 0.2 and 1.0°C (Figure 1c) caused by an albedo reduction of 0.15–0.20 in the feedback simulation relative to the nonfeedback simulation (Figure 5c). The albedo reduction and the resulting temperature feedback is of similar magnitude in spring (MAM) (results not shown), but is slightly weaker since spring is also associated with additional snowmelt and an opposing evapotranspiration feedback.

Figure 7.

CO2 fertilization and increased temperatures lead to an increased forest fraction in the Scandinavian Mountains and a decline in understory grasses in the feedback experiment.

[29] In central and eastern Europe the evapotranspiration feedback dominates (Figure 6). In winter a tendency to a negative temperature feedback mediated by increased evapotranspiration is shown, particularly in northern Spain and western France (Figure 1c). Warming temperatures and increased leaf cover in spring amplify the feedback, which expands over large parts of the central and eastern parts of Europe by the summer months (Figure 2c). This counteracts the temperature enhancement by 0.2–0.5°C (Figure 3c) compared to the nonfeedback simulation.

[30] In southern Europe a positive temperature feedback mediated by reduced evapotranspiration becomes apparent in spring, persisting in summer (JJA) (Figure 2c). Increased temperatures dry out soils by the summer in both the feedback and nonfeedback simulation. With little summer rainfall in this Mediterranean climate, summer evapotranspiration declines (Figure 6b). The negative response of vegetation cover and leaf area to drier soils results in an even stronger reduction in evapotranspiration feedback in the feedback simulation, resulting in an additional temperature increase of 0.2–1.0°C in summer (Figure 3c) compared to the simulation with prescribed vegetation. In autumn, feedbacks decline.

4. Discussion

4.1. Vegetation Response to Simulated Climate Change

[31] The relatively larger winter-temperature increases simulated over northern Europe are associated with extensive reductions in the period of snow-lie, and a complementary increase in the general length of the vegetative growing season. According to the simulations, Mediterranean-climate areas face the risk of increased dryness associated both with reduced annual precipitation and a greater evaporative demand as the atmosphere warms. These factors will result in contrasting responses in vegetation activity in northern and southern Europe.

[32] In the north, climate-induced warming positively affects annual net primary production for both forest and open land. The biochemical stimulation of photosynthesis at higher CO2 concentrations (“CO2 fertilization” [Hickler et al., 2008]) further accentuates this general trend according to RCA-GUESS. Particularly woody PFTs profit from the CO2 fertilization. The simulated shifts in vegetation zones are a combined response to migration of the isotherms that correspond to bioclimatic limits (coldest month temperatures and growing season heat sums) for the establishment or survival of individual PFTs, the transitional dynamics that follow from the colonization of new areas that become climatologically available, and changes in competitive balance between PFTs in response to changing growth conditions. Similar complex responses have been documented for (palaeo-)historical plant species shifts; the suggested mechanisms involved are linked to complex interactions between differences in CO2 response and allocation strategies between woody and herbaceous plants [Polley et al., 2002; Walker et al., 2006]. In our simulations, the most dramatic changes in vegetation structure were predicted to occur in the Scandes mountains, where first deciduous, then evergreen trees colonize extant tundra regions, heavily suppressing understory vegetation by shading (Figure 7). In central parts of Europe, the simulated vegetation changes were more subtle, involving a greater proportional abundance of deciduous PFTs in mixed forests (Figure 8). The main mechanism for this in the model is probably the improved competitiveness of the winter-deciduous habit relative to evergreenness under a longer growing season and higher CO2 concentrations. An additional explanatory factor is that some areas become accessible to shade-tolerant deciduous species as isotherms migrate northward, whereafter population and individual growth gradually increase the newcomers’ share of the overall vegetation, at the expense of the resident conifers. For example, deciduous PFTs expand northward into Russia where conditions become more favorable. The boreal forests face a stronger competition from the invasion of these PFTs, leading to a reduced evergreen fraction.

Figure 8.

CO2 fertilization and increased temperatures lead to a greater LAI and a greater mixture of PFTs in the central to eastern parts of Europe in the feedback experiment.

[33] In the southern and Mediterranean parts of Europe, warming and reduced rainfall impact negatively on the vegetation through reduced growing-season soil water availability (Figure 9).

Figure 9.

CO2 fertilization and increased temperatures lead to a reduced LAI on open land tiles during summer (June, July, and August) in southern Europe. Results are from feedback experiment and denote the change in LAI from (1961–1990) to (2071–2100).

4.2. Vegetation Feedbacks on Simulated Climate

[34] In terms of vegetation-climate feedbacks our results are in line with previous studies using GCMs coupled to simplified vegetation models, that highlight albedo-mediated warming enhancement in northern regions [Bonan et al., 1992; Levis et al., 1999] and hydrological cycling feedbacks in more southerly areas. Our results are unique in providing a high-resolution picture of how these contrasting feedback mechanisms might be distributed across Europe under the future climate, and confirm their applicability under the more complex system dynamics of a regional climate model coupled to a detailed vegetation model. Our results also illustrate the sensitivity of hydrological feedback mechanisms, in magnitude but also in sign, on the net effect of climate change on vegetation growth. The contrasting evapotranspiration-mediated feedbacks in central and eastern Europe, on the one hand, and southern Europe, on the other, reflect the relative changes in vigor, growth, leaf area and phenology of vegetation in these two “hot spot” areas. To a lesser extent, they also reflect differences in vegetation composition in these areas in terms of growth form (woody/herbaceous), life history strategies and phenology within a forest.

[35] The vegetation feedback contribution from changes in albedo and evapotranspiration is evident in temperature. However, neither variation in cloudiness nor precipitation could be attributed to vegetation dynamics in our study (results not shown). For the European domain, Atlantic convection strongly determines precipitation and cloud formation. This may tend to overwhelm any feedback signal from vegetation-mediated changes in evapotranspiration. Also, the ratio between sensible and latent heat exerts strong local control on temperature, but effects on cloud formation and precipitation will take place at the site of condensation, further away and higher up in the atmosphere, diffusing the signal. Since the vegetation-climate feedbacks vary by season the annual average effect of the contrasting feedbacks tends to cancel out (Table 2), at least in terms of the albedo versus evapotranspiration feedback in the Scandinavian Mountains. Likewise, the vegetation-mediated exacerbation of summertime evapotranspiration in southern Europe is balanced by reversal of the same mechanism in other seasons. In central and eastern Europe, however, vegetation changes amplify evapotranspiration in all seasons, however with varying strength, resulting in an overall negative feedback on temperature on an annual average.

Table 2. Seasonal Variations and Annual Average in Surface Temperature Anomalies in Degrees Celsius for the Analyzed Hot Spots and the Dominant Feedback Metrica
 DJFMAMJJASONAnnual
  • a

    α, albedo; E, evapotranspiration.

Scandinavian Mountains0.2–1 α0.2–1 α−0.2 to −0.5 E−0.2 to −0.5 E-
Central Europe−0.2 to −0.5 E−0.2 to −1 E−0.2 to −0.5 E-−0.2 to −0.5 E
Eastern Europe-−0.2 to −1 E−0.2 to −1 E−0.2 to −0.5 E−0.2 to −0.5 E
The Mediterranean0 to −0.5 E0 to −1 E0.2–1 E--

4.2.1. Albedo Feedback in the Scandinavian Mountains

[36] The boreal tree-line advance and the associated albedo reduction and temperature increase in the Scandinavian Mountains agree well with many other studies of future warming [e.g., Bonan et al., 1992; Levis et al., 1999; Bergengren et al., 2001]. The herbaceous PFT is replaced first by deciduous broad-leaved trees and subsequently evergreen conifers. The corresponding vegetation elements in the real-world ecosystems of this area constitute an altitudinal replacement series: tundra vegetation (comprising various forbs, grasses and dwarf shrubs), the tree-line-forming mountain birch (Betula pubescens ssp. tortuosa), and, at lower altitudes, the boreal conifers Scots pine (Pinus sylvestris) and Norway spruce (Picea abies). The simulated successional order of the PFTs also reflects life history strategies of pioneer versus successional tree species [Loehle, 2000], but in this case the succession is mainly driven by amelioration of winter minimum temperatures that govern the elevational limits of the respective PFTs. The albedo feedback is accentuated during winter and spring as the expanding forest cover masks snow, but the overall effect on temperature would seem to be weaker than suggested by other studies [e.g., Betts, 2000].

[37] In summer and autumn the temperature feedback becomes reversed, which is linked to the cooling effect of a greater evapotranspiration associated with the forest expansion. Forests capture a larger proportion of the incoming rainfall as interception compared to tundra, while the greater roughness of the forest canopy increases turbulent mixing, promoting the escape of the evaporated water to the open atmosphere. This results in a shift toward a greater release of latent heat (cooling) at the expense of sensible heat (warming). The energy balance reveals that spring, when snowmelt is taking place, is associated with an increased fraction of net radiation. Since the radiation absorbed is split into two components, in terms of sensible and latent heat, the latter might reduce the effect of the albedo on temperature. Our results indicate that this is the case, since a pronounced evapotranspiration feedback from the increased forest fraction is evident also in spring. Although evapotranspiration should be negligible during winter at high latitudes owing to low temperatures, the feedback is visible but yet weaker also in winter, suggesting that with future warming the evapotranspiration feedback may offset the albedo feedback in winter and spring.

[38] The simulated vegetation shifts in the Scandinavian Mountains remained in a transient state by the end of the simulation in 2100. This suggests that continued warming, or even climatic stasis after 2100, should lead to further succession toward boreal conifer forest as the long-term steady state (climax) vegetation. If so, this suggests that the positive albedo feedback on climate might grow stronger for a number of decades after 2100 until the climax vegetation is reached. By the 2071–2100 time slice, the vegetation of these areas includes a relatively significant fraction of deciduous broad-leaved trees. The same explanation may apply to the weaker Siberian albedo feedbacks identified by Göttel et al. [2008], who used the same vegetation model as in the present study in offline mode to explore the effects of updated vegetation fields on the climate simulated by the REMO RCM under a future scenario.

[39] Similar vegetation-climate feedbacks to those simulated for the Scandinavian Mountains might be expected in the Alps, but this was not found to be the case. A weaker albedo decline in the Alps was not strong enough to produce a temperature difference between the feedback and nonfeedback simulations in this area. The explanation for this probably lies in the comparatively coarse grid resolution of the study. The Alps are geologically younger than the Scandes and are characterized by steep and high topography, with the result that individual mountain ridges and peaks cover a much smaller fraction of the 50 × 50 km grid cell they intersect. As a result, the prescribed elevation and simulated climate of most grid points in the Alps more closely reflect conditions in the forested valleys than the sparsely vegetated mountain tops. The southerly location of the Alps also increases the counteracting feedback of evapotranspiration changes on temperature, relative to the effect of a marginally reduced albedo.

4.2.2. Hydrological Cycle Feedbacks in Central, Eastern, and Mediterranean Europe

[40] The relatively significant CO2-fertilization effect on woody PFT growth and cover is evident over all Europe in our simulations. Although southerly parts of Europe are characterized by drier conditions in the future, the relative PFT abundances shift slightly toward woody drought tolerant PFTs. However, in the most southern parts of Europe, increased climate-driven summer dryness has a negative impact on plant production and consequently LAI and vegetation cover, which in turn reinforces dryness in this region. Increases in drought-tolerant Mediterranean conifers (MNE) and broad-leaved evergreen trees (BE) result in a negative evapotranspiration feedback in winter for, for example, northern Spain and western France. In spring, this effect is amplified over central and eastern Europe in conjunction with leaf onset in the deciduous vegetation. In central and eastern Europe the evapotranspiration feedback remains negative (reduced warming) through the summer. In these regions the increased LAI mitigates climate driven future warming.

[41] In summer dryness in the most southern parts is heavily manifested. Most importantly, this causes a significant reduction in open land (grass) LAI, which in turn amplifies the reduction in evapotranspiration. In the nonfeedback simulation, the generally high LAI enables a higher rate of evapotranspiration, which continues until plant-available soil water is depleted. The moderate increase of woody PFTs in southern Europe is too weak to mitigate the decline in evapotranspiration under the hotter and drier future summers simulated over southern Europe. Risks for increased drought conditions in southern Europe under the future climate have been pinpointed in earlier studies [Lehner et al., 2006; Morales et al., 2007]. The positive vegetation-mediated feedbacks identified in the present study lend additional support to this hypothesis. Our results for the future climate indicate a greater seasonal amplitude in LAI in this region compared to present conditions. Mediterranean-climate vegetation exhibits a main active period during spring, when radiation and temperature conditions are conducive to growth, but soils remain well hydrated following winter precipitation [Mooney et al., 1975]. In our simulations, future warming may enable significant growth also during winter. Therefore the annual trend in, for example, LAI, NPP is slightly positive from 1961 to 1990 to 2071–2100 as a consequence of the CO2 fertilization effect, which on an annual basis offsets the negative impacts of drier summers. Similar results have been obtained in other modeling studies [e.g., Osborne et al., 2000].

4.3. Limitations

[42] Although our results point toward significant and regionally contrasting feedbacks of vegetation changes on seasonal and regional patterns of climate change, our results indicate that the overall influence of such feedbacks on climate trends over Europe may be minor in magnitude, compared with changes in greenhouse forcing inferred to result from changes in ecosystem carbon sinks/sources in a number of studies [Cox et al., 2000; Friedlingstein et al., 2006]. Changes in vegetation of the type we have investigated (i.e., “natural” ecosystem responses to changes in climatic drivers + CO2) are also much smaller than assumed, for example, by Betts [2000] and Bala et al. [2007] in their studies of albedo feedbacks in Northern high-latitude ecosystems.

[43] However, we have not considered changes in land use (e.g., afforestation/deforestation) or land management (e.g., changed species choice in forestry), which could modify the strength of vegetation feedbacks to climate. Neither have we considered any realistic crop phenology in agricultural areas. Arable lands, though still accounting for a considerable fraction of the land cover over continental Europe, have declined steadily throughout the past century. A continued contraction of croplands in favor of forest and abandoned lands (which generally develop into forest) as suggested, for example, by Kankaanpää and Carter [2004] and Rounsevell et al. [2006], would impact the strength of the investigated feedbacks relative to our study. Likely effects would include a more pronounced albedo-mediated warming feedback for northern Europe as suggested by Betts [2000], but possibly also an enhancement of the evapotranspiration-mediated feedback simulated over central and eastern Europe in our study. The latter would tend to mitigate warming, assuming that the forests maintain a greater evapotranspiration rate than agricultural land in these areas, as simulated by RCA-GUESS. The predominant albedo feedback in our results could also be modified by land management in terms of forest thinning, since thinning directly would influence the forest stand LAIs. Such effects of thinning on albedo were shown by Vesala et al. [2005]. Although biogeochemical feedbacks are excluded from the present study our results demonstrate similar ecosystem responses to CO2 forcing as other studies [e.g., Morales et al., 2007], suggesting a future increase in tree biomass and consequently also an increase in the terrestrial carbon sink, which in isolation from the biophysical feedbacks we identified has a mitigating effect on climate warming.

[44] Since European weather/climate is strongly driven by general circulation and effects of the Atlantic sea, the contribution of possible vegetation-mediated changes in evapotranspiration to cloud cover and precipitation would tend to be small compared with incoming humidity from west. Nevertheless, as seen in our results, evapotranspiration feedbacks on near-surface temperatures could be significant locally.

[45] Certain land surface processes, such as stomatal regulation and soil hydrology, are represented separately and using somewhat different formulations in both RCA and LPJ-GUESS. Although overall consistency between the models, for example with respect to simulated stomatal conductance/surface resistance, has been checked for and confirmed, such process duplication increases complexity in the model and may exacerbate uncertainty. Any bias in the models may propagate to biases also in related processes. The “offline” version of RCA3, for example, exhibits positive bias in winter temperatures in Scandinavia [Kjellström et al., 2005], which potentially may cause stronger or weaker responses in both albedo and evapotranspiration responses. Uncertainties also arise from the choice of GCM providing boundary conditions at the lateral boundaries of the RCM-simulated domain. Studies with offline impact models suggest that this can be one of the most important sources of uncertainty in the estimation of future climate-driven adjustments in terrestrial carbon storage [e.g., Morales et al., 2007]. Viewed together, the limitations in our study underline the importance of joint benchmarking efforts among ESM groups to develop suitable tools, data sets and protocols for the characterization and evaluation of vegetation feedbacks, providing a basis for the eventual incorporation of vegetation dynamics as a standard component in climate models.

5. Conclusion

[46] Here we have identified three European hot spots of future biophysical vegetation-climate feedbacks. We have demonstrated the role of vegetation driven reductions in the albedo in northern Europe as an amplifier of climate warming and the role of hydrological cycling feedbacks having to opposing effects for southern and central Europe, mediated in part by the underlying response of vegetation to climatic forcing. Further simulations incorporating potential future trends in land use would most likely modify the strength of the investigated feedbacks, probably amplifying their importance assuming a future continuation of recent historical land use trends. Our results serve to emphasize the importance of accounting for vegetation and land cover changes for a more realistic representation of future climate change, particularly when the focus is on regional and seasonal aspects of change.

[47] An additional important aspect to consider in the case of the Scandinavian Mountains is the fact that the predicted positive feedback from dynamic vegetation changes (increased tree cover, growth and LAI) to climate (warmer temperatures) could not have been fully characterized with preexisting models or methods, for example, global Earth System Models with dynamic vegetation and biophysical feedbacks, or iterative coupling [cf. Göttel et al., 2008].

[48] The lack of knowledge about advanced statistical tools for addressing uncertainties in the growing number of both regional and global ESMs complicates the task of quantifying vegetation-climate feedbacks realistically and ruling out effects from biases in the models. From our perspective this is one of the main issues to address in order to make ESM models robust tools in future climate modeling.

Acknowledgments

[49] Anna Wramneby was funded in part by a grant to B. Smith from the Swedish Research Council. B. Smith acknowledges support from the Swedish Research Council for Environment, Agricultural Science, and Spatial Planning. This study is a contribution to the Mistra Swedish Research Programme for Climate, Impacts and Adaptation and the Lund University research program Modeling the Regional and Global Earth System (MERGE). The model simulations were performed on the climate computing resource Tornado funded with a grant from the Knut and Alice Wallenberg foundation and housed at the National Supercomputing Centre at Linköping University.

Ancillary