Studying ecosystem stability to global change across spatial and trophic scales

Get Complete Project Material File(s) Now! »

Studying ecosystem stability to global change across spatial and trophic scales


Many ecosystems around the globe are threatened by changes in climate and land use, which impact biodiversity at different levels. Mountain ecosystems, in particular, are especially vulnerable to the effects of climate change, as they harbour many species that are near their environmental tolerance limits. Changes in climate drive species range shifts and impact physiological processes, and might also impact the provisioning of ecosystem services (Bellard et al. 2012). Land-use changes, be it by conversion of natural habitats into agricultural or urban lands, or by abandonment of managed areas, could aggravate the effects of climate change, as well as contribute to large and sudden changes of available habitats and ecosystem services (Asner et al. 2004). Climate, however, is not only predicted to change in its long-term average, but also with regard to extreme events (e.g. drought), expected to intensify in many regions (IPCC 2012).
Drought affects plant reproduction, growth and survival and can ultimately lead to changes in forest (Park Williams et al. 2012) and grassland (Gu et al. 2007) productivity, to changes in vegetation composition of landscapes (Clark et al. 2016), and result in significant forest dieback at the global scale (Allen et al. 2010). Forest dieback can have cascading effects on biodiversity, carbon, water and nutrient cycling, and ultimately on the provisioning of ecosystem services, such as carbon uptake and storage (Anderegg et al. 2013). Such effects are likely very important in forest–grassland ecotones of mountain environments, where many tree species live close to their lower temperature limits and may reach their soil moisture limits in dry valleys (Goldblum & Rigg 2010). This is the case even in regions like the European Alps, where tree lines are further constrained by land use for farmlands, grazing and mowing (Carlson et al. 2014). In recent years, drought events have caused Swiss forests to suffer significant diebacks of Scots pine (Pinus sylvestris L.), favouring replacement colonization by pubescent oak (Quercus pubescens L.) and a turnover of forest composition (Rigling et al. 2013). Moreover, plant communities in mountainous areas are threatened by changes in land use that affect plant community structure and composition (Tasser & Tappeiner 2002). For example, simulations of vegetation dynamics in the European Alps predict that land-use abandonment and climate warming will interact and increase forest expansion towards higher elevations (Dirnböck et al. 2003; Boulangeat et al. 2014a).
Although forest–grassland ecotones in the European Alps are facing environmental changes originating from three fronts (i.e. gradual and extreme climate change, and land-use abandonment), so far no study has investigated their joint effects, especially from a forecasting perspective (Seidl et al. 2011). While climate change and land abandonment should lead to forest expansion and woody encroachment of subalpine and alpine pastures (Asner et al. 2004), drought stress increases tree mortality, causing forest dieback. Consequently, we can expect that drought might counter the effects of gradual climate change and land-use abandonment, but this is likely to depend on drought frequency and intensity, as well as on the identity of species within communities. Milder droughts may contribute to faster woody encroachment of subalpine and alpine pastures by favouring species adapted to drier environments. On the other hand, very severe and, or frequent droughts are likely to slow forest progression. Such effects may not be homogeneous in space, especially if certain areas are prone to more intense or more frequent drought (Dobbertin et al. 2005; Worrall et al. 2013).
Here, we study, in a spatially explicit manner, how drought frequency and intensity interact with climate and land-use practices and affect forest–grassland ecotones in the European Alps using the landscape dynamic vegetation model FATE-HD (Boulangeat et al. 2014b). Although FATE-HD does not simulate drought effects at the individual and physiological level, it can capture drought at the community level providing useful insights for management and conservation planning of complex ecosystems. For example, ecosystem management and conservation in the European Alps focuses on maintaining a bundle of ecosystem services (Grêt-Regamey et al. 2008; EC 2015), managing for a high diversity of habitats and on protecting biodiversity per se. Although this includes maintaining forest cover, there is also an important focus on avoiding woody encroachment of open habitats. FATE-HD provides information on these different conservation goals.
Specifically, we explored i) under which conditions drought reversed the trend of forest expansion that is observed under climate change and land-use abandonment; ii) whether forest–grassland ecotones suffered important changes in taxonomic and functional diversity when exposed to extreme events; and iii) the possible spatio-temporal dynamics of these changes. Finally, iv) we evaluated the consequences of drought regimes in the context of current land-use management and the provisioning of ecosystem services in the European Alps.

Materials and methods

Study area

We focused our study on the forest–grassland ecotone habitats of the Ecrins National Park (ENP), situated in south-east France in the French Alps. The park covers an area of 178 400 ha (elevation ranging from 669 to 4102 m a.s.l.), with a rich diversity of plant species (ca. 2000) and ecosystems, from mountainous to alpine habitats – the majority being open habitats (60% of the park surface). Land use consists mainly of agricultural activities (grazing, 48%; crop fields and mown grasslands, 9 8%; and forest management, 14%), which are accurately mapped (Esterni et al. 2006).

The base model: FATE-HD

FATE-HD has already been parameterized to explore the synergistic effects of land-use (LU) and climate changes (CC) on the vegetation of the ENP (Boulangeat et al. 2014a), and we have now extended it to incorporate drought effects. We first give a brief description of the base model (further details in the section FATE-HD ‘base model’ description in Appendix 2 and in Boulangeat et al. 2014b) and then follow with a more detailed description of the new drought module.
FATE-HD models the spatio-temporal dynamics of plant functional groups (PFGs) by explicitly simulating their population dynamics and dispersal, interactions for light resources, and their responses to climate and different LU regimes. FATE-HD has been parameterized for 24 PFGs representative of both the taxonomic and functional diversity of the rich flora in the ENP (Boulangeat et al. 2012). They consist of six chamaephyte groups (C1-6), 10 herbaceous groups (H1-10) and eight phanerophyte groups (P1-8), each occupying up to five height strata and passing through four ages (1–4) that have different responses to disturbances (see Table S1 in Appendix 2). The abundance of a given stratum in a pixel determines the amount of light that reaches lower strata. Interactions for light resources are simulated by accounting for the amount of light reaching each PFG cohort in a stand and the PFG’s light preferences. Responses to climate are simulated through habitat suitability (HS) maps (constructed a priori based on observed occurrences) for each PFG, and climate change is simulated by changing HS maps at regular intervals (see FATE-HD ‘base model’ description in Appendix 2). Land-use disturbances are modelled in a spatially explicit manner, by assigning mowing, grazing (with intensities low, medium or high) or no disturbance to each pixel (see FATE-HD ‘base model’ description in Appendix 2). Model output consists of yearly strata and PFG abundances per pixel.

Simulating drought events

Whether or not a PFG was affected by drought depended on the comparison of the PFG’s past drought exposure to simulated yearly drought intensity values. To calculate each PFG’s past drought exposure, we combined PFG occurrences (from the vegetation data base of the Conservatoire Botanique National Alpin; see Boulangeat et al. 2012 and CBNA 2015) with monthly values of a moisture index (MI; Thornthwaite 1948) across the entire French Alps for 1961–1990 (see Parameterising and simulating drought effects in Appendix 2). MI was calculated as the difference between precipitation and potential evapotranspiration (negative MI indicating drought; see Parameterising and simulating drought effects in Appendix 2). For each PFG, we extracted the distribution of monthly MI values from each plot where it occurred (hereafter MI1961–1990). We then defined drought intensity (Din) as the lowest MI value in a year for each occurrence plot. Finally, a PFG’s past drought exposure was defined as the distribution of these experienced Din values (hereafter Din1961–1990).
At each year, the PFGs’ past drought exposure calculated above was compared with values of Din within each pixel (Din maps; Fig. 4). The comparison triggered, or not, consequences of drought events through two sequential modelling steps: i) ‘identifying drought effects’ and ii) ‘modelling drought response’.

Identifying drought effects under past and future conditions

In the simulations, yearly Din values per pixel were obtained from past observations of MI values in the French Alps (validation runs) or from calculations of MI values using future climate predictions (future drought scenario runs). Drought was detected for a PFG in a given pixel, for a given simulation year, when the pixel Din was ‘abnormally’ low relative to the PFGs past drought exposure (Din1961–1990 distribution, Fig. S1 in Appendix 2). The drought status was classified as ‘no drought’, ‘moderate drought’ or ‘severe drought’, depending on two ‘drought detection thresholds’, which were defined as deviations from mean values of PFGs’ past drought exposure (Din1961–1990 distribution). No drought was detected if the pixel Din value was greater than x̅ – 1.5 x SD of Din1961–1990 (x̅ and SD being the mean and standard deviation, respectively; step 2.1 in  effects (step 2) are evaluated within a pixel j. If HSij or Dinij are below reference values (HSref and x̅ 1.5SD of Din161-1990, respectively) PFG fecundity and recruitment are set to 0 (medium -grey arrows and boxes). Additionally, if Dinij crosses the reference value, one drought year is added to the PFG’s cumulative drought effects counter (thin dashed arrows and light grey box). Severe drought effects occur if conditions 2.3.1.ii or 2.3.2 are met (dark grey arrows and boxes), consisting in immediate and post-drought effects (full and dash-dot arrows, respectively). Otherwise, only moderate drought effects are caused (2.1 and 2.3.1.i; medium grey full arrows). Drought recovery is simulated by subtracting one (phanerophytes and shrub chamaephytes, C4) or two drought events from the cumulative drought effects counter (thin, dark grey and dashed arrow). Small light grey squares indicate the ‘drought sensitivity’ parameter and the total number of squares indicates the size of the counter (‘cumulative drought response’ parameter). See Table S3 in Appendix 2 for full parameter list and refer to main text for further details.
Effects of severe drought depended on the accumulation of past drought events (‘cumulative effect of drought’; steps 2.3.1 ii and 2.3.2). This cumulative effect of drought was meant to simulate the fact that after long periods of water stress, less intense droughts may also have severe effects on tree mortality (McDowell et al. 2008; Allen et al. 2010). Cumulative effects of drought were twofold and are regulated by two PFG-specific parameters: the PFG’s sensitivity to a severe drought (‘drought sensitivity’; Table S3 in Appendix 2) and the PFG’s response to successive droughts (‘cumulative drought response’; Table S3 in Appendix 2). Drought sensitivity expressed the number of droughts a PFG must experience before suffering severe effects due to a severe drought (step 2.3; see ‘Modelling drought responses’ below). For herbaceous groups, severe drought effects occurred during the first severe drought they experienced (drought sensitivity = 1). In contrast, chamaephytes and phanerophytes were less sensitive: chamaephytes only suffered severe effects during the second drought they experienced, while phanerophytes and shrubs (C4) were only affected severely during the third drought event (step 2.3.1 ii; years represented as light grey squares in Fig. 4). After a certain number of droughts experienced, the PFG’s tolerance is weakened and all drought events (moderate or severe) have severe consequences. The cumulative drought response parameter expressed how many successive drought events were tolerated by a PFG before any subsequent drought events started having severe effects (step 2.3.2). Again, chamaephytes and phanerophytes (together with group C4) were more tolerant and only suffered severe effects from subsequent droughts after three and five drought events, respectively (total number of squares in Fig. 4). Herbaceous groups only needed two drought years to be severely affected by any subsequent drought.
Finally, to simulate drought recovery (and avoid accumulating drought years indefinitely), we removed one drought year (phanerophytes and shrub chamaephytes) or 2 years (herbaceous groups and most chamaephytes) from the PFGs’ cumulative effect counters during each non-drought year (dark grey dashed arrow in Fig. 4).

READ  Participatory rural appraisal: a methodology to facilitate roots-driven change in rural areas 

Modelling drought responses

Drought effects were twofold, immediate and/or post-drought (occurring the year after drought), in order to simulate demographic responses during and after drought occurs (Allen et al. 2010). Drought immediately affected a PFG’s recruitment and fecundity, which were set to 0 during the present drought year (moderate and severe droughts). A severe drought also increased PFGs mortality (0– 60% depending on the PFG type, soil moisture preference and age) and caused PFGs to resprout (0–80% depending on PFG type, soil moisture preference and age). Post-drought effects were only modelled after a severe drought. To this end, PFG recruitment and fecundity were set to 0, PFG mortality was increased, and resprouting of PFGs was activated (although less than for immediate drought effects: 0–20% for mortality and 0– 50% for resprouting). We calculated each PFG’s soil moisture preference class (0 = drought tolerant to 3 = drought intolerant; see details in Parameterising and simulating drought effects in Appendix 2) considering both PFGs’ past drought exposure (via their MI1961–1990 distributions) and expert knowledge on the soil moisture requirements of the PFGs. Drought-intolerant PFGs responded with higher mortality rates, and, across PFGs, younger and older PFGs (extremes of the size gradient) also suffered higher mortality rates (McDowell et al. 2008). Herbaceous and most chamaephyte PFGs never resprouted during drought, did not suffer post-drought mortality, but always resprouted after a severe drought. Younger individuals (age 1) were never capable of resprouting (see Parameterising and simulating drought effects in Appendix 2 for further details, and Table S3 for the full list of drought-related parameters).
Finally, we also simulated the protective effect of canopy cover as a buffer against drought effects. Canopy cover has been shown to increase seedling survival, by an amelioration of local microclimate conditions in terms of air and soil temperature, radiation and humidity (Gómez-Aparicio et al. 2005; Kane et al. 2011). Hence, in simulations, pixel Din values were increased by 25% in pixels where tree cover (strata > 1.5 m) was at least 40% (Esterni et al. 2006) – recall that less negative Din values correspond to less severe drought.

Simulation experiments

Simulations started with an initialization phase of 850 years during which current climate and land-use regimes were modelled (Fig. 5). This phase allowed a stabilization of PFGs and achieving the current vegetation state before any future scenarios of climate, land-use or drought regime changes were applied (Boulangeat et al. 2014b). All scenario simulations started from the last year of the initialization phase, which will be referred to as year 0 hereafter.
We simulated one scenario of gradual CC, five scenarios of increasing drought frequency that were combined with three scenarios of increasing drought intensity (see below) and with two scenarios of LU change, totalling to 30 scenarios runs (Fig. 5). Since changes in drought regimes are thought to be a consequence of CC, we always changed background climate through its impact on habitat suitability for PFGs. Additionally, we ran two baseline simulations with only gradual CC (i.e. no drought), each with a scenario of LU.
Drought was implemented similarly to CC, by feeding maps of drought intensity (Din) values. Like maps of ‘current’ HS, ‘current’ Din maps were calculated by averaging past MI values across years 1961–1990. Since drought events are caused by extreme values of temperature and/or precipitation (IPCC 2012), we used the predicted temperature and precipitation maps for 2080 (following the A1B scenario described in FATE-HD ‘base model’ description in Appendix 2) to calculate future maps of Din values. Current IPCC predictions indicate that drought frequency and intensity are to increase in the future (IPCC 2012); hence, we simulated three different drought intensities with linearly increasing drought frequencies and fixed periods without drought events to test our hypotheses. This allowed the vegetation to recover by avoiding long periods of continuous drought if frequency was high. Drought was then set to occur either every year or every 2, 4, 8 or 16 years (five drought frequency scenarios), with a 10-year no-drought period after each sequence of five drought events.
Future and current Din maps were alternated to create drought and no-drought years, respectively. As for drought intensity, we calculated three levels of intensity (‘low’, ‘medium’ and ‘high’) that would not greatly deviate from climate predictions. Medium intensity corresponded to forecasted Din values for the year 2080, and low/high intensity corresponded to an increase/decrease of these values by 20%, respectively (three intensity scenarios; see Fig. S1 in Appendix 2).
Figure 5. Simulation experiment workflow. An initialisation phase (0-850 years) allowed reproducing the current vegetation state of the Ecrins National Park. The last 50 years of the initialisation phase (800-850 years) were used to validate the drought module, while the last year (year 850) was the starting point of scenario simulations, which lasted 200 years. Climate change was implemented from years 15 to 90. Land-use changes were implemented at year 4, remaining unchanged until the end of the simulations. Drought regimes were initiated at year 15, lasting at least as long as climate change, up to year 105.
Land-use scenarios consisted of removing all grazing and mowing activities to simulate LU abandonment, or to continue current LU practices to simulate a ‘business-as-usual’ scenario (two LU scenarios). Land-use abandonment was applied at year 4 until the end of the simulation, whereas gradual CC was applied by changing PFG habitat suitability at regular 15-year intervals, starting at year 15 until year 90 (see FATE-HD ‘base model’ description in Appendix 2 and Boulangeat et al. 2014a). The duration of drought regimes depended on the scenario of drought frequency, but covered at least the period of gradual CC (years 15–90) and finished before year 105. For all scenarios, simulations were run for a total of 200 years and replicated three times (Fig. 4).
We validated the drought module by running a simulation using observed data of drought events that occurred between 1961 and 1990 (Fig. 4). Model output was compared to vegetation plots from the ENP database and the previous validation of FATE-HD (Boulangeat et al. 2014b). We found that the simulated vegetation represented the observed vegetation of the park well and concluded that the inclusion of drought effects led to good model performance (see Validation of drought module in Appendix 2).

Analysis of results

To answer our first question regarding the effects of drought and LU on forest expansion, we analysed how different combinations of drought intensity, drought frequency and LU regimes influenced the speed at which forest and shrubland migrated towards higher elevations. Forest and shrubland pixels were identified based on the percentage of tree cover (strata > 1.5 m), which was larger than 60% for forest and between 10% and 60% for shrubland (Esterni et al. 2006). Rates of forest and shrubland expansion (RFE and RSE, respectively) were estimated independently for three different time frames. The first reflected the initial impacts of gradual CC, LU and drought disturbances (years 0–49). The second reflected responses to on-going CC and drought events and medium-term responses after they ended (years 50–149). The last time frame reflected long-term responses to gradual climate and LU changes, as well as recovery from drought events and the eventual establishment of new equilibria (years 150–200). Hence, for each time frame (and each scenario), we regressed yearly maximum elevation obtained for forest and for shrubland pixels against time to obtain the rates of expansion (regression slopes). Responses of RFE and of RSE to drought and LU regimes, and their interactions, were analysed separately for each time frame using analyses of variance (ANOVAs). Four levels of drought intensity (‘no drought’, ‘low’, ‘medium’ and ‘high’) and six of drought frequency (‘no drought’, every year and every 2, 4, 8 and 16 years) were used as independent factors. Land use was used as a factor with two levels (‘abandonment’ and ‘business-as-usual’). Model selection was based on Akaike Information Criterion (AIC) values, model parsimony and analyses of residuals.
Taxonomic response of the forest–grassland ecotone to simulated drought, CC and LU, was assessed by quantifying PFG turnover both spatially and temporally, using a measure of β-diversity. The ecotone was spatially delimited for each scenario at year 0, using a buffer distance around the upper tree line (1000 m above and 500 m below), tree line being defined at the third quartile of elevation values of forest pixels. Mean -diversity was calculated using a multiplicative decomposition of – and -diversity, calculated as the inverse Simpson concentration (Whittaker 1972): where p is the relative abundance of each PFG across pixels (for -diversity), or in each pixel i (for -diversity), and n the total number of communities. Alpha and -diversity are bounded between unity and the maximum number of PFGs; -diversity is bounded between unity and the maximum number of communities (Tuomisto 2010). For temporal turnover, -diversity was calculated per pixel (i.e. communities) with reference to year 0 at subsequent 5-year intervals (year five against year zero, year 10 against year zero, etc.), then averaged across all ecotone pixels to obtain a value per pair of years. Spatial turnover was calculated across all ecotone pixels (for a given year) every 5 years.
We also explored how community-averaged soil moisture preference changed spatially under different drought regimes and different LU practices. Focusing again on the ecotone, we calculated community-weighted mean values of soil moisture preference classes (CWMSM) every 5 years, by weighing PFG soil moisture preference values (SMj) by PFG relative abundances (abundj, j being a PFG) for each pixel (Garnier et al. 2004; Violle et al. 2007): CWM SM SM j abund j (Eq. 3)
Since we were interested in mapping increases or decreases of CWMSM, rather than following its temporal evolution, we calculated changes in CWMSM per pixel, as the difference between CMWSM values of a given year and year 0. Negative values indicated shifts towards communities with preference for drier soils, while positive values indicated shifts to communities with higher moisture requirements. Resulting CWMSM changes were mapped for different scenarios of drought intensity/frequency and LU management to obtain a spatial image of functional community shifts. In addition, we assessed whether changes towards communities with higher or lower soil moisture preference were linked to elevation, by calculating Pearson product–moment correlation coefficients between values of change and elevation across ecotone pixels.


When considering all scenarios of drought and land-use regimes, rates of forest expansion (RFE) towards higher elevations were not significantly different from the rates of shrubland expansion (RSE; Fig. S4 in Appendix 2). While RSE was always significantly affected by drought intensity and frequency, RFE only responded significantly to these factors during the early phases of the simulations (years 0–49 and 50–149 in Fig. 6a and Tables S5 and S6). Both RSE and RFE were more affected by different drought intensities, rather than frequencies (Table S6 in Appendix 2). Also, RSE and RFE responded differently to the interaction between drought intensity and frequency, which significantly affected RFE between years 0 and 149, but only had a significant effect on RSE during the last 50 years.

Table of contents :

Appendix 1
Appendix 2: Supplementary materials to Chapter I
Appendix 3: Supplementary materials to Chapter II
Appendix 4: Supplementary materials to Chapter III
Appendix 5: Supplementary materials to Chapter IV
Appendix 6: Simulating plant invasion dynamics in mountain ecosystems under global change
Curriculum Vitae


Related Posts