Dryland ecosystems are a major source of land cover, account
for about 40% of Earth's terrestrial surface and net primary
productivity, and house more than 30 % of the human population. These
ecosystems are subject to climate extremes (e.g. large-scale droughts and
extreme floods) that are projected to increase in frequency and severity
under most future climate scenarios. In this modelling study we assessed the
impact of single years of extreme (high or low) rainfall on dryland
vegetation in the Sahel. The magnitude and legacy of these impacts were
quantified on both the plant functional type and the ecosystem levels. In
order to understand the impact of differences in the rainfall distribution
over the year, these rainfall anomalies were driven by changing either
rainfall intensity, event frequency or rainy-season length. The
Lund–Potsdam–Jena General Ecosystem Simulator (LPJ-GUESS) dynamic vegetation
model was parameterized to represent dryland plant functional types (PFTs)
and was validated against flux tower measurements across the Sahel.
Different scenarios of extreme rainfall were derived from existing Sahel
rainfall products and applied during a single year of the model simulation
timeline. Herbaceous vegetation responded immediately to the different
scenarios, while woody vegetation had a weaker and slower response,
integrating precipitation changes over a longer timeframe. An increased
season length had a larger impact than increased intensity or frequency,
while impacts of decreased rainfall scenarios were strong and independent of
the season characteristics. Soil control on surface water balance explains
these contrasts between the scenarios. None of the applied disturbances
caused a permanent vegetation shift in the simulations. Dryland ecosystems
are known to play a dominant role in the trend and variability of the global
terrestrial CO
Dryland ecosystems account for about 40 % of Earth's terrestrial surface
and net primary productivity (Grace et al., 2006; Wang et al., 2012) and
shelter more than 30 % of the human population (Gilbert, 2011). These
ecosystems are subject to climate extremes that are projected to increase in
frequency and severity under most future climate scenarios (IPCC, 2014;
Sillmann et al., 2013). Such extremes (e.g. large extent droughts and extreme
floods) can have a devastating impact on the ecosystems and livelihoods of
global drylands, as well as amplify pressure on fragile economic
structures (Ibrahim, 1988; United Nations Office for the Coordination of
Humanitarian Affairs, 2013). The Sahel, situated south of the Sahara
desert, is one of the largest dryland areas of the world, covering more than
The Sahel is mostly dominated by savanna grasslands. These complex biomes
consist of a sparse cover of C
Although the vegetation structure and ecosystem productivity in water-limited ecosystems is mainly driven by annual total precipitation (Lehmann et al., 2014; Sankaran et al., 2005, 2008), intra-annual rainfall variability, which is characterized by the variability in rain event intensity, frequency and timing of the wet season, has a large impact on the vegetation as well, by changing the spatial and temporal availability of soil water for plant uptake (Berry and Kulmatiski, 2017; Case and Staver, 2018; Guan et al., 2018; Kulmatiski and Beard, 2013; Xu et al., 2018; Zhang et al., 2018, 2019). The year-to-year variation in these characteristics is significant in global drylands, including the Sahel (Reynolds et al., 2007; Zhang et al., 2017). Climate projections for the end of the 21st century generally show a delay in timing of the rainy season, with average shifts of around 5 to 10 d for the Sahel (Dunning et al., 2018; IPCC, 2014; Pascale et al., 2016). Total precipitation is expected to decrease in the western parts and to increase in the central and eastern parts of the Sahel, although a high variability remains among the different climate model predictions (Biasutti, 2019; Pascale et al., 2016). Furthermore, an increase in rain event intensity, coupled with a decrease in frequency, has been observed in recent years (Panthou et al., 2014; Taylor et al., 2017) and is projected for the coming century (Dunning et al., 2018). Even though the region has a long history of adapting to drastic changes in rainfall (Mortimore, 2010), it is still uncertain how current and future changes in rainfall regimes will impact the plant functional responses in the Sahel and in drylands in general.
Dryland vegetation is known to respond in contrasting ways to intra-annual
rainfall variability. An increased frequency of heavy rainfall events is
reported to facilitate woody encroachment in savanna ecosystems (Kulmatiski
and Beard, 2013; Zhang et al., 2019), but this response is modulated by the
underlying soil texture, as a more intense rainfall leads to a lower tree
cover on soils with a finer texture (Case and Staver, 2018). Other studies
found that regions with a given amount of total seasonal rainfall have a
higher woody cover under a more frequent but less intense rainfall
climatology, which can be explained by differentiated tree and grass water
use strategies (Good and Caylor, 2011; Xu et al., 2015, 2018). D'Onofrio et
al. (2019) found a positive relationship between grass cover and rain event
frequency but only a weak link between tree cover and rainfall seasonality
characteristics for drylands (MAP
In order to gain a more detailed process-based insight into how dryland vegetation is affected by the distribution of rainfall over the rainy season, we used a dynamic vegetation model to study the impact and legacy of single anomalous rainy seasons on the vegetation. The approach presented here is therefore complementary to earlier studies, such as Guan et al. (2018), which mainly assessed the impact on the vegetation of long-term changes in intra-seasonal rainfall variability, informing on the ecosystem state under prolonged changes in rainfall regime. Hence, the vegetation response in such studies is subject to cumulative effects of repeated rainfall disturbances, obscuring the underlying mechanisms that drive these responses.
We aimed at assessing the impact of different rainfall scenarios on the vegetation response at four flux tower sites across the Sahel (Tagesson et al., 2016; Table 1), investigating the response of individual plant functional types (PFTs) and of the ecosystem as a whole. We parameterized the Lund–Potsdam–Jena General Ecosystem Simulator (LPJ-GUESS) dynamic global vegetation model (Smith et al., 2014) for the Dahra site in Senegal (Tagesson et al., 2015), using field measurements and a literature study. The model was evaluated at all Sahel sites by testing whether it significantly improved the representation of the site ecosystem fluxes relative to the published version of the model (Smith et al., 2014). The model experiments were set up as a disturbance event, where we altered the rainfall during 1 year in the meteorological driver time series. We changed the total rainfall together with one of the underlying seasonal characteristics (i.e. intensity, frequency or length), while keeping the other two characteristics invariant.
Adopting this approach, we addressed the following research questions: (1) how do years of extreme rainfall with different seasonal characteristics impact the fluxes and composition of dryland ecosystems in the Sahel in the period following the extreme event, and (2) how do the magnitude and legacy of these impacts vary across the different plant functional types?
Overview of the different flux tower sites used in this
study, together with the 1979–2016 mean annual precipitation and its
standard deviation (MAP, mm yr
The Sahel is a semi-arid ecoclimatic transition zone, bridging the Sahara
desert in the north with the Sudanian savanna in the south. It is usually
defined by the 150 and 700 mm isohyets delineating its northern and
southern borders, respectively. In this study we used data from four flux
tower sites that have been established in the Sahel, measuring
land–atmosphere carbon, water and energy exchanges, together with
meteorological data (Tagesson et al., 2016; Table 1). The flux towers are
located at Dahra in Senegal (DAH), Agoufou in Mali (AGG), Wankama in Niger
(WFF) and Demokeya in Sudan (DEM) (Fig. 1). All sites consist of a grassy
savanna with a sparse tree cover, growing on sandy arenosol soils. Annual
total rainfall varies from 339 mm in the west (Dahra) to 164 mm in the east
(Demokeya), with mean annual temperatures around 29
Map including the Sahel region, showing the locations of the different flux tower sites: Dahra (Senegal), Agoufou (Mali), Wankama (Niger) and Demokeya (Sudan). Imagery © 2020 TerraMetrics; map data © 2020 Google.
We used the LPJ-GUESS process-based dynamic global vegetation model, which
simulates the global vegetation structure with its associated carbon,
nitrogen and water cycles (Smith et al., 2014). Similar to many global
models, LPJ-GUESS uses plant functional types to represent physiological,
morphological and phenological differences in vegetation. Out of the 12
standard PFTs in LPJ-GUESS, three are relevant for the Sahel: tropical
broadleaved evergreen trees, tropical broadleaved deciduous trees and C
Phenology of the drought deciduous PFTs is based on a water stress scalar in the model. Low values of this scalar represent stress due to reduced soil water content, leading to a reduction of photosynthesis through stomatal closure. When this variable drops below a given threshold, the dry season starts and deciduous trees will shed their leaves. Likewise, when this scalar rises above this threshold new leaves will be produced, taking into account a prescribed minimum dormancy period (Smith et al., 2014).
Soil hydrology is represented by a two-layer bucket model with percolation between the layers and drainage at the bottom (Gerten et al., 2004). The upper layer has a depth of 0.5 m, while the lower layer is 1 m deep, adding up to a total soil depth of 1.5 m. Rainfall will replenish plant-available water in the upper layer up to field capacity, above which excess water will be expelled as surface runoff. The lower soil layer is supplied with water by percolation from the upper layer. Transpiration by plant canopies will in turn reduce the water content in both soil layers. Different PFTs can have different root biomass distributions across the soil layers; e.g. grasses will have 90 % of their root biomass in the upper layer, while trees have deeper roots in the model (Table 2). LPJ-GUESS has previously been used in sub-Saharan Africa and other savanna studies, and the model is known to give a reasonable representation of large-scale sensitivities to drought in drylands at the global scale and for Africa specifically (Ahlstrom et al., 2015; Baudena et al., 2015; Boke-Olén et al., 2018; Brandt et al., 2017, 2018; Lehsten et al., 2016). Nonetheless, the parameterization of the PFTs has never been optimized for the drylands in the Sahel specifically.
Important PFT parameter values used in LPJ-GUESS:
photosynthetic pathway (Photo), specific leaf area (SLA, m
We adjusted the parameterization of LPJ-GUESS to the local conditions by
updating two plant functional traits (specific leaf area and wood density)
to values from Nielsen (2016) and Sibret (2021). The tropical evergreen
tree PFT was based on
The model was evaluated against flux tower data from the four Sahel sites by comparing a 10 d moving average of the measured daily net ecosystem productivity (NEP) and evapotranspiration (ET) time series with model predictions. Model performance metrics were summarized in a Taylor diagram (Taylor, 2001).
By default, LPJ-GUESS is driven by daily interpolations of monthly Climatic
Research Unit and National Centers for Environmental Prediction (CRU–NCEP)
meteorological forcing data (Viovy, 2018). To improve the temporal
resolution of the meteorological forcing, we used meteorological data
extracted from the WATCH (WATer and global CHange) Forcing Data methodology applied to ERA-Interim
reanalysis (WFDEI; Weedon et al., 2014) with substituted Multi-Source
Weighted-Ensemble Precipitation v1.2 data (MSWEP; Beck et al., 2017, 2019).
Both reanalysis datasets contain daily averages of meteorological data from
1979 to 2016 (Fig. 2). The data have a 0.5
Median and variability of the rainy-season
characteristics for the Sahel sites (Table 1) studied:
Overview of the general simulation timeline for each scenario. During the experiment period, 1 single year of the meteorological time series is disturbed, as illustrated by the different branches, and immediately followed by an ensemble of spin-down runs, consisting of average rainfall years only. Each horizontal segment represents a cycle of meteorological forcing data.
The rainfall disturbance experiments were developed to depict an increase or
decrease of the total precipitation in a given year by 2 standard deviations of the annual rainfall, therefore representative of
extreme years in the historical time series. This disturbance was applied in
such a way that only one of the three seasonal characteristics (intensity,
frequency or length; Table 3) changed, while the other two remained
invariant, thus creating a target rainy season for the selected year (Table 4). A detailed description of the used algorithms can be found in the
Supplement, Sect. S1. In order to preserve the internal meteorological
consistency with the other drivers (air temperature and incoming short-wave
radiation), we resampled all data from the original meteorological drivers:
for each DOY (day of the year) in the goal scenario, we found a date with matching rainfall
(
Definitions of the different rainy-season characteristics used in this study.
The six simulation scenarios (Table 4) were applied to each year of the
meteorological cycle that had a total rainfall close (
Overview of the scenarios with actual rainy-season characteristic values for the Dahra site (average and standard deviation taken over all ensemble members).
This leads to an internally consistent set of meteorological model drivers
for all six disturbance scenarios (Table 4), each simulated by an ensemble
of
For each site and each scenario, the impact of the disturbance and its legacy on vegetation were finally quantified as the difference between the output of the reference and disturbance runs, displayed as a function of time since the disturbance event, and finally averaged over all ensemble members (Fig. 4). Impact legacy is calculated as the last year for which the average impact is larger than its standard deviation, i.e. when the uncertainty on the impact becomes larger than its difference with the reference run, resulting in a relatively conservative estimation of legacy. We analysed the response of individual PFTs, as well as the ecosystem as a whole, by quantifying the impact on leaf area index (LAI), carbon cycling and surface water balance. We show the full impact time series for the Dahra site as an illustration, together with a summarized result to compare key response descriptors (maximum impact and legacy) between the different sites. Full impact time series for all sites can be found in the Supplement.
Concept of the model experimental set-up, consisting of a disturbance simulation in which a rainfall disturbance is applied, and a reference run, which is based on the same meteorological drivers but without any disturbance applied. Vegetation impact is described by maximum impact (amplitude) and legacy (years). The average impact is derived by subtracting the output of the reference simulations from the output of the disturbance simulations for each ensemble member and then taking the average of the result over all ensemble members. The legacy is calculated as the last year for which the average impact is larger than its standard deviation.
The updated parameterization of LPJ-GUESS captures the net ecosystem
productivity (NEP) and evapotranspiration (ET) that were measured at the
Dahra flux tower site, to which the model was parameterized (Fig. 5). Carbon
uptake follows the timing of the rainy season, but the amplitude of both NEP
and ET are underestimated over the whole time series. The uncertainty on the
Dahra flux tower NEP measurements varies around an average of 1.5 gC m
Time series of a 10 d moving average of
From all applied scenarios, an increase in rainy-season length caused the
largest increase in LAI at all sites and for all PFTs. Especially C
Taylor diagrams showing the correspondence between
modelled and observed daily NEP values (10 d moving averages) for the
Sahel flux tower sites, evaluated over
Scenarios of increased rain event frequency generally had a slightly larger impact than those of increased intensity. The impact of both scenarios was the weakest for deciduous trees (15 %–25 %) and slightly stronger for evergreen trees (25 %–45 %) and grasses (30 %–55 %), with an exception for the Demokeya site, which showed a larger impact of these scenarios on the evergreen trees (65 % for increased intensity and 85 % for increased frequency) (Fig. 8).
Response of the vegetation to the different rainfall
scenarios for the Dahra site, as a function of years since the disturbance
event.
Summarized overview of the vegetation response to the different rainfall scenarios for all sites, showing the maximum impact (%) on the LAI for each PFT, along with its legacy (years), which is defined as the last year for which the impact is 1 standard deviation away from the reference value. Sites are represented by different colours; scenarios are represented by symbols.
In contrast, for the scenarios of decreased total rainfall, the simulated
reduction in LAI was largely independent of the rainy-season characteristic
that was adjusted, although scenarios of decreased intensity depicted a
slightly lower decline in LAI overall (Figs. 7 and 8). For C
For all scenarios and all sites, grasses responded immediately to changes in
precipitation, with the highest impact occurring during the perturbed year
(Figs. 7d and S5–S7). In contrast, the tree PFTs exhibited their peak impact in
the year following the disturbance (Figs. 7e–f and S5–S7). At all sites,
deciduous trees experienced a reversal in response (“overshoot”) following
the initial impact, which is up to the same order of magnitude as the
initial impact, in particular for scenarios of increased rainy-season length
and rainfall reduction (Figs. 7f and S5–S7). Depending on the site and scenario,
these overshoots can last for multiple (
On the ecosystem level, the reference net ecosystem productivity (NEP) was
mostly positive but relatively small, ranging between
Impact of the different scenarios on the cumulative NEP
at the Dahra site.
Summarized overview of the net ecosystem flux response to the different rainfall scenarios for all sites, showing the impact on the net ecosystem productivity (NEP) and the cumulative NEP. Different sites are represented by different colours, while the symbol shapes represent the applied scenarios.
The disturbance scenarios have a varying impact on the surface balance
between water evaporation, runoff and infiltration into the soil (Figs. 11 and
12). For disturbances that are based on a higher event frequency or a higher
rainfall intensity, more than half of the added rainfall will be evaporated
or lost to runoff for all sites so that the resulting amount of infiltrated
water will be reduced accordingly (Fig. 12). Increased rainfall intensity
caused a
Although fire can play a major role in affecting the vegetation structure of
African savannas, sites with a MAP below 350 mm are more rarely regulated by
fire because of the low fuel availability (Sankaran et al., 2008). All
sites considered in this study have a MAP of 339 mm or less (Table 1), and
most fire events are anthropogenic. Nevertheless, as annual precipitation
levels may increase under future climate scenarios, fires may play an
increasing role at these sites in the future. The standard LPJ-GUESS model
has a relatively simple fire module, where ignition is based on fuel load
and litter moisture (Thonicke et al., 2001). In our study, fires only
occurred at the wettest sites (Dahra and Wankama) when the fuel load was
high and desiccated during a dry period following an occasional wet year. In
these cases, fires mostly contributed for a small fraction (
Impact of the different disturbance scenarios on surface
water balance. Reference values and impact on
Impact of the different disturbance scenarios on surface
water balance for all sites, showing the maximum impact on the surface
evaporation, runoff and percolation as a percentage of added/reduced
rainfall amount (%
A combination of the updated model parameter values and daily-scale
WFDEI–MSWEP meteorological drivers improved the agreement between model
simulations and flux tower measurements of NEP for all Sahel sites. Our
model disturbance experiments illustrated a strong contrast between
scenarios of increased and decreased rainfall for all sites. For increased
rainfall scenarios, the impact strongly depends on how rainfall is
distributed over the season, while this was not the case for decreased
rainfall scenarios. Out of the rainfall addition scenarios, increasing the
length of the rainy season had the strongest impact at all sites, especially
on C
The improved model performance was expected, as the published model PFTs (Smith et al., 2014) represent generic tropical species, while the new parameter values are specific for dryland ecosystems. A one-at-a-time sensitivity analysis revealed that updating other relevant parameters did not lead to a significant improvement in simulating the carbon fluxes and ET (not shown), but parameter covariance sensitivities are still to be tested. Moreover, the daily MSWEP data show a better match with precipitation measured at the flux tower (Fig. S3) and capture the onset and end of the rainy season better than the interpolated CRU–NCEP monthly data. One factor that is not included in these simulations is livestock grazing, which was found to have a positive impact on both gross primary productivity (GPP) and ecosystem respiration (ER) (Tagesson et al., 2016b), although further studies are needed to fully understand this mechanism. At the Dahra site, cattle density was the highest during 2010, potentially explaining the higher discrepancy with the model that year (Fig. 5) (Tagesson et al., 2016b). Simulated reference values of surface runoff are relatively high when compared against earlier published ranges for the Sahel (Fekete et al., 2002). This has been observed in earlier land surface model intercomparison studies as well, stressing the need for a good representation of soil hydrology in vegetation models (Grippa et al., 2017).
At all sites, the model simulates co-existence of woody and herbaceous PFTs,
which will compete for resources and therefore generate complex vegetation
dynamics. The herbaceous layer generally responded more strongly and swiftly
to perturbations in precipitation than the woody vegetation for almost all
sites and scenarios, especially to increases in rainy-season length (Figs. 7–8). This contrasting behaviour reflects differences in plant
representation in the model. Increased precipitation will lead to increased
carbon uptake, which for grasses can be allocated to roots and leaves only.
The tree PFTs will need to allocate a significant amount of carbon to woody
components as well, which in turn will provide a safety net during the
scenarios of decreased rainfall. The difference in the timing of the impact
between grasses and trees is also partly due to the differences in root
distribution. Grasses will be mostly affected by the water content in the
upper soil layer, as it contains 90 % of their root biomass in the model.
Therefore, they will directly respond to changes in precipitation (Brandt et
al., 2018; Gherardi and Sala, 2015). In contrast, trees have 40 % of
their root biomass in the lower soil layer of the model, where the water
content integrates changes in precipitation over a longer timeframe.
Together with physiological differences (e.g. allocation to woody parts),
this explains the longer reaction time of the trees in the model. Further
differences in response between evergreen and deciduous trees are due to
their difference in SLA and phenology. Positive disturbances of increased
rainfall initially benefit both woody PFTs, but the positive impacts last
longer on evergreen trees, while the positive impacts on deciduous trees are
followed by a negative overshoot, especially for the scenario of increased
season length. Similarly, evergreen trees recover more slowly from negative
disturbances than deciduous trees, which display a positive overshoot
following the initial negative impact. These results show that single-year
disturbances can shift the weights in the competition for resources among
the different PFTs for several years. Kulmatiski and Beard (2013) have
shown experimentally that an increase in rainfall intensity (without
changing the total rainfall) will increase aboveground woody plant growth
and decrease grass growth. This behaviour is not observed in our model
study. However, as Kulmatiski and Beard (2013) argued, this increase
reflects the ability of woody plants to increase their rooting depth, a
process that is not included in our model, which only simulates two soil
layers for which total root biomass can vary but in which the PFT root
distribution between the two layers remains fixed. Earlier research showed
that the water use of
On the ecosystem scale there is an increase in carbon uptake in response to
a year of increased precipitation, but most of this gain is quickly lost
again during the following years (Figs. 9 and S4). The largest part of the
photosynthesized carbon will be allocated to leaf and root biomass of the C
The contrast between the scenarios of increased rainfall was also simulated
by Guan et al. (2018) for scenarios with long-term changes in
precipitation, although this is only found for regions of higher mean annual
precipitation (700–1600 mm yr
It is expected that variations between the different sites are largely due to differences in historical meteorological conditions, as all other model parameters remained the same across sites. Variations between the sites were most clearly distinguished in the scenarios of reduced rainfall, where especially the impact legacy varied across the sites. No clear relationship was found between site conditions (Table 1) and ecosystem response to any of the disturbances. However, sites with a lower MAP, such as Demokeya and Agoufou, experienced a lower NEP impact and a shorter legacy from scenarios of reduced rainfall, compared to the wetter sites. The impact is likely lower because these drier sites will conceive a lower reference vegetation cover, leading to a lower impact on heterotrophic respiration after an exceptionally dry year. Similarly, wetter sites had a higher fraction of rainfall that percolated into the soil than drier sites, while surface evaporation increased more at drier sites than wetter sites. This may be due to shading, which is higher in wetter sites because of a higher vegetation cover. Nevertheless, in order to derive a clear relationship between site conditions (e.g. MAP) and disturbance impacts, a follow-up study could focus on sites along a stronger gradient in site conditions (e.g. a north–south precipitation gradient).
Finally, our results seem to contrast earlier research which has shown that phenology of cropland and grassland in sub-Saharan Africa is mainly driven by photoperiodicity, while in our model a longer rainy season will cause a longer growing season (Adole et al., 2019). Photoperiodicity is only implemented for crop PFTs in LPJ-GUESS, while for the natural vegetation PFTs that were used in this study, simulated phenology is driven by water availability and therefore follows the rainy season. However, at the local scale the importance of photoperiodicity is diminished in the Sahel. While individuals of several species are photoperiodic, phenological plasticity is strong, and a longer rainy season does seem to bring a longer growing season due to cohort and species succession.
The approach developed in this study presents a unique way to investigate
the impact of different rainy-season characteristics on the vegetation in
the Sahel. Our algorithm allows us to create artificial rainfall scenarios
which strongly resemble the original rainfall data, while also retaining the
internal consistency with other meteorological variables. Some ensemble
members in our scenarios may suffer from a loss of autocorrelation in
temperature or incoming radiation, as the algorithm for constructing the
artificial scenarios occasionally had to consult neighbouring pixels or
longer time periods. However, as these dryland ecosystems are mostly
sensitive to rainfall and as it was rainfall which was varied the strongest
(
As this is a modelling study, its outcome is as reliable as the model assumptions and parameterization, in addition to the quality of the meteorological drivers. Allocation to carbon pools happens at the end of each simulated year, which may influence our results, although we mainly look at inter-annual impacts. A model version which includes daily carbon allocation for grasses has been developed for studying grass dynamics and grazing potential in more detail, which could be further developed to include daily allocation for tree PFTs as well (Boke-Olén et al., 2018). The representation of drought stress and hydraulic dynamics through the soil–plant–atmosphere continuum is expected to play an important role in determining the impact of drought response of ecosystems, especially in drylands (Medlyn et al., 2016). However, like many vegetation models, LPJ-GUESS has a relatively simple representation of these processes, based on empirical relationships (Gerten et al., 2004; Smith et al., 2014). Recently, efforts have been made to improve these processes in the Ecosystem Demography (ED) model, by including a representation of the hydraulic pathway through the plant, connecting phenology with hydraulic status, and by parameterizing the hydraulic model based on plant hydraulic traits (Xu et al., 2016). Adapting these or similar ideas for LPJ-GUESS will most likely improve both the validation and the reliability of the results presented in this study. The quality of soil hydrology representation in the model may have an influence on the results of this study as well, given the importance of runoff and percolation for the vegetation impact. Furthermore, implementing a photoperiodicity-driven phenology may be necessary to upscale this research to the regional level (Adole et al., 2019). Finally, although fires play a major role in regulating woody cover in African savannas, its impact was limited for the dryland sites in this study, and the model uncertainty was high where fires occurred. Most likely fire will play a larger role at lower latitudes, where the MAP levels are sufficient to generate the necessary fuel load for fires to occur. Studying the impact of rainfall disturbances on fire occurrence in those regions will lead to a better understanding of the complex disturbance-driven dynamics of mesic savannas (Sankaran et al., 2008), especially when LPJ-GUESS is coupled with more sophisticated fire models such as SPITFIRE (SPread and InTensity of FIRE), which significantly improve the fire model performance in those regions (Thonicke et al., 2010).
The LPJ-GUESS dynamic vegetation model is
available upon request at
The supplement related to this article is available online at:
WV, GS, SH and HV designed the research. WV performed model experiments and analysed the data. JA, PNB, BC, JD, RF, LK, TS and TT collected the field data. WV drafted the paper, and all authors contributed to writing the paper.
The authors declare that they have no conflict of interest.
The authors would like to sincerely thank Marie Combe for the fruitful discussions and Moustapha Mbaye for the local support in Senegal.
This research was part of the U-TURN (Understanding Turning Points in Dryland Ecosystem Functioning) project, which was supported by the Belgian Federal Science Policy Office (BELSPO) (grant no. SR/00/339) in the framework of the STEREO III (Support to Exploitation and Research in Earth Observation) programme.
This paper was edited by Akihiko Ito and reviewed by two anonymous referees.