Projection of vegetation impacts on future droughts over West Africa using a coupled RegCM-CLM-CN-DV

This study investigates the projected effect of vegetation feedback on drought conditions in West Africa using a regional climate model coupled with the National Center for Atmospheric Research Community Land Model, carbon–nitrogen (CN) module, and dynamic vegetation (DV) module (RegCM4.3.4-CLM-CN-DV). The role of vegetation feedback is examined based on simulations with and without the DV module. Simulations from four different global climate models are used as lateral boundary conditions for historical and future periods (historical: 1981–2000; future: 2081–2100). Using the standardized precipitation evapotranspiration index (SPEI), we quantify the frequency, duration, and intensity of droughts over two focal regions of the Sahel and the Guinea Coast in West Africa. With the dynamic vegetation (DV) considered, future droughts are expected to become more prolonged and enhanced over the Sahel. The drought is defined when the SPEI value is lower than the threshold (− 1) in the simulations with DV than those without DV. Over the Guinea Coast, the impact of DV is opposite compared with that over the Sahel. Additionally, we show that the simulated annual leaf area index associates well with annual minimum SPEI with the correlation coefficient up to 0.91, particularly over the Sahel, which is a transition zone, where land–atmosphere feedback is relatively strong. Results signify the importance of vegetation dynamics in predicting future droughts in West Africa, where the biosphere and atmosphere interactions play a significant role in the regional climate regime.

Keywords: West Africa; SPEI; LAI; Vegetation feedback; Drought

Supplementary Information The online version of this article (https://doi.org/10.1007/s10584-020-02879-z) contains supplementary material, which is available to authorized users.

West Africa is extremely vulnerable to climate change and projecting its future climate is a challenging task (Cook et al. [14]). From the 1970s, a long period of drought was observed over West Africa, lasting until the late 1990s. There is no clear consensus about whether the future of the West African hydroclimate will be drier or wetter. Some studies projected drying trends (Hulme et al. [30]), whereas others predicted a wetter future (Hoerling et al. [27]; Kamga et al. [32]; Maynard et al. [41]). Caminade and Terray ([9]) examined the simulated rainfall over the Sahel at the end of twenty-first century with 21 models from the Coupled Model Intercomparison Project (CMIP) Phase 3 (CMIP3). They argued that different model projections are highly uncertain because future rainfall may be affected by changes in surface conditions (e.g., vegetation, land use, and soil moisture) that have not been considered in CMIP3 models. Roehrig et al. ([54]) combined the CMIP3 and CMIP Phase 5 (CMIP5) global climate models (GCM) and found that the western Sahel shows a drying trend, whereas the eastern Sahel shows an opposite trend.

Limited-area models, i.e., regional climate models (RCMs), which are forced with lateral boundary conditions (LBCs) derived from GCMs, are often used as they can capture finer-resolution details, compared with GCMs (Kumar et al. [35]) since the physics of RCMs dominate the signals imposed by large-scale forcings (i.e., LBCs from GCMs). Akinsanola and Zhou ([2]) investigated projected changes in extreme summer rainfall events over West Africa with data from the Coordinated Regional Climate Downscaling Experiment (CORDEX) models. Results showed the RCMs participating in CORDEX reasonably reproduced the observed pattern of extreme rainfall over the region. Future projections under the representative concentration pathways showed a statistically significant decrease in total rainfall and an increase in the consecutive dry days and extreme rainfall. In another study, Akinsanola and Zhou ([3]) projected future changes in summer monsoon rainfall under the RCP8.5 scenario using the Rossby Centre Regional Climate Model (RCA4), driven by six CMIP5 GCM scenarios. The authors found that summer monsoon rainfall in a future period (2074–2099) is projected to increase than that in present-day period (1980–2005) over most parts of West Africa. In addition, Dixon et al. ([19]) studied the changes in monsoon precipitation over the Sahel region of West Africa under a global sea surface temperature (SST) increase by 4 K using a 2D-model by Peyrillé et al. ([53]). The results indicated that direct forcing would shift the precipitation band toward the south, whereas forcing associated with a global increase in SST would reduce precipitation.

Because climate and greenhouse gas concentrations continuously change, changes are expected in the vegetation as well (Yu et al. [74]). Thus, global and regional climate models should incorporate more representative and reliable prognostic vegetation dynamics instead of prescribing vegetation composition and structure, particularly for the regions where biosphere–atmosphere interactions are important (Alo and Wang [4]; Patricola and Cook [51]; Wramneby et al. [70]; Xue et al. [72]; Zhang et al. [79]). Charney et al. ([10]) first suggested that precipitation could change dynamically in response to vegetation changes, and suggested that the changes in precipitation over the Sahel were due to reduction in vegetation and increase in albedo. Various studies documented biosphere–atmosphere interactions (Kim and Wang [34]; Patricola and Cook [50]; Wang and Eltahir [66]), but relatively few RCM studies considered prognostic vegetation dynamics, (Cook and Vizy [13]; Garnaud et al. [22]; Wang et al. [67]; Yu et al. [75]).

Cook and Vizy ([13]) developed a vegetation model coupled with an RCM to estimate the influence of global warming on South America by allowing interactions between climate and vegetation. With the simulation of the future climate under the CMIP3 A2 scenario, the authors found a reduction in vegetation cover of approximately 70% in the Amazon rainforest along with a widespread increase in grass and shrubland in the east by the end of twenty-first century. This highlights the importance of considering vegetation dynamics in RCMs. Garnaud et al. ([22]) combined the Canadian Regional Climate Model (CRCM5) with the Canadian Territorial Ecosystem Model to investigate the impact of a vegetation model for simulating the present-day climate over North America. The results showed that including vegetation dynamics improved the model's performance in some regions while introducing new biases in other regions, owing to biases in the simulated leaf area index (LAI, a dimensionless quantity defined as the green leaf area per unit ground surface area). This atmosphere-vegetation interaction also introduced long-term memory, which was estimated using a lagged correlation between temperature/precipitation and LAI. Wu et al. ([71]) utilized RCA-GUESS (Smith et al. [56]), which is a regional earth system model coupled with a dynamic vegetation model, and investigated the role of vegetation dynamics in climate of Africa under the RCP8.5 projected climate scenario. The authors showed that including vegetation processes amplified the warming trend and enhanced precipitation reduction over rainforest areas, which highlights the impact of including vegetation processes in a climate model.

Recently, Wang et al. ([67]) introduced dynamic vegitation (DV) into the International Center for Theoretical Physics Regional Climate Model (RegCM4.3.4, Giorgi et al. [24]) with carbon–nitrogen (CN) and DV schemes (RegCM-CLM-CN-DV) of the community land model (CLM4.5, Lawrence et al. [36]; Oleson et al. [46]), and validated the coupled model over tropical Africa. With the RegCM-CLM-CN-DV, Yu et al. ([75]) and Erfanian et al. ([21]) examined the impacts of vegetation dynamics on the climate using multiple LBCs from past and future GCM simulations over West Africa. Yu et al. ([75]) showed that climate impact of dynamic vegetation feedback was found mainly in semiarid areas of West Africa with little signal in the wet tropics. Similarly, Erfanian et al. ([21]) also demonstrated substantial sensitivity of the simulated precipitation, evapotranspiration, and soil moisture to vegetation representation. Including DV in the model eliminates potential inconsistencies between prescribed vegetation and climate, but it can cause climate drift and enhance model biases (Erfanian et al. [21]).

While many studies focused on projecting future climate over West Africa, a limited number of studies quantitatively assessed drought frequency, duration, and intensity, as well as considering the role of vegetation dynamics on the drought processes in this region. In view of the uncertain future of drought-prone regions, this study aims to understand the impacts of vegetation feedback on the future of droughts over West Africa. Specifically, the standardized precipitation evapotranspiration index (SPEI) is used to depict the vegetation feedbacks on drought characteristics according to frequency, intensity, and duration over West Africa. Following previous studies with RegCM-CLM-CN-DV (Wang et al. [67]; Yu et al. [75]; Erfanian et al. [21]), we examined the drought characteristics simulated with and without vegetation dynamics for the historical and future periods and showed the signals of DV on the drought characteristics in two selected regions of West Africa.

This study uses state-of-the-art RegCM-CLM-CN-DV (Wang et al. [67]). Specifically, RegCM4.3.4 (Giorgi et al. [24]) and CLM4.5 (Lawrence et al. [36]; Oleson et al. [46]) with CN dynamics and DV are coupled to simulate various atmospheric, land, biogeochemical, vegetation phenology, and vegetation distribution processes. RegCM is a regional model that uses an Arakawa B-grid finite differencing algorithm along with a terrain-following σ-pressure vertical coordinate system. The model uses the Grell et al. ([26]) dynamics from the hydrostatic version of the Pennsylvania State University Mesoscale Model version 5, and the radiation scheme from the Community Climate Model (Kiehl et al. [33]). The model includes four different convection parameterization schemes, namely (1) the modified-Kuo scheme (Anthes et al. [5]), (2) Tiedtke scheme (Tiedtke [62]), (3) Grell scheme (Grell [25]), and (4) Emanuel scheme (Emanuel [20]) along with the non-local boundary layer scheme of Holtslag et al. ([28]). The cloud and precipitation schemes come from Pal et al. ([48]). The aerosols algorithm follows Solmon et al. ([58]) and Zakey et al. ([76]).

To solve the various land processes in the model (e.g., surface biogeochemical and biogeophysical processes, ecosystem dynamics, and hydrological processes), CLM4.5 considers 15 soil layers, 16 distinct plant functional types (PFT), and up to 5 snow layers in each grid cell (Lawrence et al. [36]). Optional components available in this model are the CN and DV modules. The CN module not only simulates CN cycles and plant phenology but also estimates vegetation height, stem area index, and LAI. The DV module projects the fractional coverage of different PFTs and corresponding temporal variations at yearly time steps based on carbon budget from CN. If CN and DV modules are inactive, the vegetation distribution and composition in the model are prescribed according to the observed data sets (i.e., static vegetation, hereafter referred to as SV).

A total of 18 different numerical simulations are used in this study, as listed in Table 1, with 2 evaluation runs and 16 experimental runs with the climate change scenarios. Numerical simulations are conducted in two distinct configurations, one in which the CN-DV module is activated (i.e., DV runs) and the other in which the CN-DV module is not activated (i.e., SV runs). Additionally, the LBCs are derived from ERA-Interim for the evaluation with SV and DV (1989–2008, i.e., EvalSV and EvalDV) and from four GCMs for the historical (1981–2000) and future (2081–2100) periods with SV and DV, i.e., HistSV and HistDV and FutSV and FutDV, respectively, under the RCP8.5 scenario. The GCMs used in this study include the Community Earth System Model (CESM) (Hurrell et al. [31]), Geophysical Fluid Dynamics Laboratory Model (GFDL) (Team et al. [60]), Model for Interdisciplinary Research on the Climate–Earth System Model (MIROC) (Watanabe et al. [69]), and Max Planck Institute Earth System Model (MPI-ESM) (Giorgetta et al. [23]). These GCMs were chosen as they can capture present-day climate over West Africa (Cook and Vizy [12]; Roehrig et al. [54]) and their climate supports a reasonable present-day vegetation according to CLM.5-CN-DV (Yu et al. [75]). The model grid is configured using a 50-km horizontal grid spacing and 18 vertical layers from the surface to 50 hPa. The model parameterizations are the same as those used in previous studies over the same region (Alo and Wang [4]; Saini et al. [55]; Wang et al. [67]), as summarized in Table 1.

Table 1 Description of 18 different RegCM-CN-DV simulation setups

Wang et al. ([67]) extensively evaluated the RegCM-CLM-CN-DV model for simulating regional climate and ecosystems in Africa. The evaluation was performed using LBCs from the ERA-Interim (1989–2008), and with and without vegetation dynamics. Yu et al. ([75]) and Erfanian et al. ([21]) also examined the impacts of vegetation dynamics on the climate and ecosystems using multiple LBCs from past and future GCM simulations. Building upon these previous studies, this study focuses on the impacts of vegetation dynamics on the regional drought characteristic (i.e., frequency, duration, and intensity) over two focal regions of the West African domain, i.e., the Sahel and the Guinea Coast (Fig. 1).

Graph: Fig. 1 Averages of precipitation (left column), air temperature (middle column), and precipitation surplus/deficit (right column) differences for summer (June, July, and August), 1981–2000, between observations (UDEL and CRU) and ensemble historical runs (HistSV and HistDV). The top and bottom black boxes show two focal regions: the Sahel and the Guinea Coast, respectively

Various drought indices, such as the Palmer Drought Severity Index (PDSI) (Palmer [49]) and Standard Precipitation Index (SPI) (McKee et al. [43]), have been used to assess drought events. Vicente-Serrano et al. ([65]) suggested the SPEI, which uses the deficit between precipitation and potential evapotranspiration (PET) and considers the impact of temperature, a major control on potential evapotranspiration. Since the development of SPEI, various drought studies have adopted this index (e.g., Boroneant et al. [8]; Deng [16]; Li et al. [37], [38]; Lorenzo-Lacruz et al. [39]; Paulo et al. [52]; Sohn et al. [57]; Spinoni et al. [59]; Yu et al. [73]). For example, McEvoy et al. ([42]) used SPEI as a drought index to monitor conditions over Nevada and Eastern California, suggesting that SPEI was a convenient tool to describe drought in arid regions. Recently, Diasso and Abiodun ([17]) investigated the future impacts of global warming and reforestation on drought patterns simulated by RCMs over West Africa using the SPEI.

In this study, we chose SPEI, considering the advantages of the index as suggested by Naumann et al. ([45]) and Um et al. ([63]). Drought indices that account for the impact of temperature, such as PSDI or SPEI, are generally preferred especially when future climate scenarios are involved (Begueria-Portugues et al. [7]). Furthermore, the SPEI can be quantified simply by fitting the difference between precipitation and PET to the log-logistic distribution, whereas PDSI can be quantified by simulating the calibrated water balance models.

We estimated the SPEI using the approach of Beguería and Vicente-Serrano ([6]). While precipitation is the simulated output of RegCM4.3.4-CLM-CN-DV, PET can be derived from the model outputs following the approach proposed by Thornthwaite ([61]). Mavromatis ([40]) showed that the use of simple or complex methods to calculate the PET provides similar results. For a given month (j) and year (i), the monthly water surplus or deficit (Di, j) is calculated by Eq. (1) given below:

Graph

where PR is the precipitation and PET is the potential evapotranspiration. Then, accumulated monthly water surplus or deficits at time scale k ( $X_{i,j}^k$ ) is calculated based on Di, j. In this study, we chose 12 months for the time scale. As suggested in Vicente-Serrano et al. ([65]), ${\mathit{\text{SPEI}}}_{i,j}^k$ is estimated by fitting $X_{i,j}^k$ to the log-logistic distribution by means of the L-moments method. In this study, a drought event is defined when the SPEI value is less than − 1.

This section briefly evaluates the model performance with observed climate and vegetation characteristics. The model evaluation with the ERA-Interim for 1989–2008 (EvalSV and EvalDV in Table 1) is documented in Wang et al. ([67]), whereas the evaluation with different GCMs for 1981–2000 (HistSV and HistDV in Table 1) can be found in Erfanian et al. ([21]). Thus, we briefly evaluate the ensemble means of HistSV and HistDV in comparison with the observational data. Relative to the observational data from the University of Delaware (UDEL) and Climate Research Unit, both HistSV and HistDV (Figs. S1 and 1) follow the observed spatial patterns of precipitation and temperature. Both overestimated precipitation over the Guinea Coast and underestimated it over the Sahel. For instance, monthly mean precipitation over the Guinea Coast in summer (June, July, and August, JJA) is found to be 206.8 mm/month for UDEL with 42.6% and 46.8% overestimation for HistSV and HistDV, respectively. In general, such wet/dry biases lead to cool/warm biases in air temperature via the enhancement/reduction of evaporative cooling in the Guinea Coast/Sahel. We also investigated the precipitation surplus/deficit that is used for calculating the SPEI values to analyze the drought characteristics. We found that the differences of both HistDV and HistSV from the observational precipitation surplus/deficit follow same pattern as witnessed for precipitation in corresponding cases. To sum up, the model performs slightly better with SV than with DV in the historical runs for most cases. However, note that DV could eliminate potential inconsistancy between prescribed vegetation and climate, particularly for the future projections. Refer to Figs. S2 and S3 for the model evaluations with the ERA-Interim (EvalSV and EvalDV), which are generally consistent with HistSV and HistDV.

With the addition of vegetation dynamics, the LAI (Fig. 2) is overestimated in the Guinea Coast, whereas it is underestimated in the Sahel (HistDV-HistSV), which is consistent with the abovementioned biases in climate conditions. Mean LAI for HistSV over the Sahel and Guinea Coast are 0.63 and 2.15, whereas those for HistDV are 0.15 and 2.61, respectively. Note that the run without vegetation dynamics (HistSV) uses the Moderate Resolution Imaging Spectroradiometer (MODIS)–based monthly varying climatological LAI values, i.e., observational data. Over the Sahel, the model underestimates the woody plants and grass coverage with a significant overestimation of bare ground area, which can be attributed to biases in the vegetation dynamics of the model (i.e., CLM-CN-DV) and physical climate of RegCM, i.e., dry bias (Wang et al. [67]; Erfanian et al. [21]). The dry/wet bias in the atmospheric forcings over the Sahel/Guinea Coast contributes to the underestimated/overestimated LAI, which then leads to additional decreases/increases in precipitation for that region (Fig. 1). Note that similar dry and wet biases were also simulated with another regional earth system model coupled with the DV model (RCA-GUESS) (Wu et al. [71]).

Graph: Fig. 2 Averages of leaf area index (LAI) from observation (MODIS), which is used for SV runs (EvalSV, HistSV, and FutSV), and LAI simulated in DV for the evaluation run and experiment ensemble runs (EavlDV, HistDV, and FutDV) in the first row and their differences in the second row. Dotted regions show the areas that pass the two-tailed significance level with α = 0.01

This section investigates the changes in climate, vegetation, and drought characteristics between the future and historical periods in the experimental runs (HistSV vs. FutSV and HistDV vs. FutDV). First, the projected changes in the climate conditions in the future, relative to the historical periods, are examined in Fig. 3, showing similar spatial patterns of changes in SV and DV. In DV (FutDV-HistDV), average differences in precipitation for JJA are approximately − 23.6 and 4.4 mm/month over Sahel and the Guinea Coast, respectively. Furthermore, atmospheric warming caused by the increased CO2 concentration in the future scenario leads to widespread increases (positive values) in temperatures by 4 to 5 °C for both SV and DV ensembles (second column of Fig. 3), as expected.

Graph: Fig. 3 Difference in precipitation (mm/month, left column), air temperature (°C, middle column), and precipitation surplus/deficit (mm/month, right column) between different experimental simulations (HistSV, HistDV, FutSV, and FutDV) for summer (June, July, and August). Dotted regions show the areas that pass the two-tailed confidence level with α = 0.01

Projected LAI changes (FutDV-HistDV of Fig. 2) result from both climate change and CO2 fertilization effects. Note that LAIs in HistSV and FutSV are identical to the observational data. In the DV ensemble, slight decreases are projected in the Sahel, while noticeable increases are projected in the Guinea Coast with an average of 1.0 m2/m2. The spatial variation of the LAI changes generally follows that of the precipitation changes (FutDV-HistDV of Fig. 3), as both are interactively simulated in the model.

In the future, the precipitation surplus/deficit shows a general decline for both SV and DV ensembles (FutSV-HistSV and FutDV-HistDV in the right column of Fig. 3). Overwhelming decreases in the precipitation surplus (i.e., increases in deficit) are found except for the southern part of the Guinea Coast, which can be attributed to future atmospheric warming (the center column of Fig. 3). In Fig. 4, we find that the changes in the precipitation surplus/deficit lead to similar changes in drought frequencies between the future and historical periods for both SV and DV ensembles, although the spatial variations in the drought frequency changes are relatively noisy. Slight increases by 0.2% and 1.9% of the frequency in the Sahel in the future period relative to the historical period are found in the SV and DV ensembles, respectively. However, an increase of 4.1% and decrease of − 4.0% in the Guinea Coast are found in the SV and DV ensembles, respectively. The impact of precipitation increase (leading to a decrease in drought frequency) dominates the impact of temperature increases in the DV ensemble (leading to an increase in drought frequency) partly because the precipitation increase is further enhanced in the DV ensemble (FutSV-HistSV vs. FutDV-HistDV in Fig. 3).

Graph: Fig. 4 Difference in drought frequencies between different experimental simulations (HistSV, HistDV, FutSV, and FutDV). Drought event is defined as an event with an SPEI value of less than − 1, relative to the whole period

Including vegetation dynamics in a land–atmosphere coupled model makes the model more complex, but is closer to the real world for climate simulation and projections. Therefore, we focus on investigating the impact of vegetation dynamics on simulated droughts (i.e., the difference between DV and SV for the future or FutDV-FutSV) in this section. Investigating the difference of LAI between DV and SV for the future period (FutDV-FutSV of Fig. 2), we find that the LAI for the DV ensemble is smaller than that of SV over the Sahel and larger over the southern part of the Guinea Coast. LAI differences between SV and DV ensembles show similar patterns both in the historical and future periods. In the historical period (HistDV vs. HistSV), LAI biases are caused by the biases from both CLM-CN-DV and RegCM as the LAI of HistSV is based on the observational data, i.e., MODIS, as previously mentioned. Yet, the lower LAI in the Sahel in the future period (FutDV vs. FutSV, average difference of − 0.51 m2/m2) is not necessarily a bias in the simulations because the future LAI in SV is assumed to be identical to historical climatological LAI as is found in the historical SV ensemble. Note that such changes in the LAI in the future do not accompany changes in the dominant vegetation types in the Sahel region. In contrast, for the Guinea Coast, the higher simulated LAI (averaged difference of 1.54 m2/m2) is associated with differences in land cover (from grasses to woody plants).

As examined in Erfanian et al. ([21]), along with lower LAI in the Sahel, when comparing DV with SV, higher albedo, lower cooling, lower evapotranspiration, and lower precipitation are simulated. Over the region below 10° N, wetter and colder climate conditions are projected with the DV ensemble compared with the SV ensemble, resulting in a greater precipitation surplus (FutDV-Fut SV in the right column of Fig. 3). Consequently, the frequencies of drought events −10% lower over the Gulf Coast and 18% higher over the Sahel, based on the ensemble averages (FutDV-FutSV). We find that such characteristics in the ensemble averages are consistently captured in each ensemble member to different extents (Fig. S4). On average, drought frequencies are higher in the Sahel, ranging from 13 to 29% among the different LBCs and lower in the Guinea Coast, ranging from − 5 to − 20% among the different LBCs.

The regional averages of SPEI over the two different regions present the impact of vegetation dynamics on simulated drought intensity and duration (Fig. 5). Ensemble SPEI averages and the black-lined boxes for the drought events, defined with the averaged SPEI values less than − 1, show that DV runs simulate more prolonged and more severe droughts for the historical and future periods over the Sahel, which is also shown by the negative values in the difference figures (i.e., HistDV-HistSV and FutDV-FutSV, last two rows of each panel). Specifically, the drought event is not identified in FutSV with the minimum SPEI of − 0.94, whereas 4 drought events across the total period of 42 months are characterized by an SPEI as low as − 1.61 in FutDV (first panel in Fig. 5). Such findings are generally opposite for the Guinea Coast, as shown by the positive values in the difference figures, with drought events being depicted in FutSV but not in FutDV. We also show that these findings are consistent with the ensemble members with four different LBCs. For instance, in MIROC, the first drought event is characterized by a period of 10 months, mean SPEI of − 1.32, and minimum SPEI of − 1.49 in HistSV, and with a period of 14 months, mean SPEI of − 1.75, and minimum SPEI of − 1.95 in HistDV. Furthermore, prolonged droughts with lengths of 64 months are projected in FutDV only. The data summarized in Table 2 clearly suggest that the impact of DV on drought frequency (i.e., the ratio of drought period relative to the whole study period) over the Guinea Coast shows the opposite comparison to that over the Sahel, where vegetation feedback clearly increased drought frequency. For instance, the drought frequency increases from 10.48% of HistSV to 27.16% of HistDV, and from 10.68% of FutSV to 29.01% of FutDV. We note that these findings are qualitatively consistent with the results of Diasso and Abiodun ([17]), which suggest longer and intense dry periods in the Sahel.

Graph: Fig. 5 Monthly SPEI averaged for two focal regions of the Sahel (left column) and the Guinea Coast (right column) in ensembles and the individual member of the experimental runs (HistSV, HistDV, FutSV, and FutDV) with different LBCs of CCSM, GFDL, MIROC, and MPI-ESM. Boxes indicate the drought events defined by an SPEI value of less than − 1

Table 2 Ensemble drought frequency (%) over the Sahel and the Guinea Coast with experimental runs (HistSV, HistDV, FutSV, and FutDV). Drought events are defined where the SPEI value is less than − 1

Finally, we perform the correlation analysis between annual maximum LAI and annual minimum SPEI over the regions for both historical and future periods (Fig. 6). Drought events are reflected in the relatively lower annual minimum SPEI because leaf growth is limited during such events. Therefore, a large portion of West Africa sees a strong positive association between the maximum LAI and minimum SPEI with DV activated. Relatively strong correlations with a coefficient up to 0.91 are found along the Sahel, which may be attributed to land–atmosphere feedbacks that are relatively strong in transition zones. Over most of the Guinea Coast, the correlation is weak or even negative, as as vegitation growth is not limited by water in the region.

Graph: Fig. 6 Spearman's rank correlation coefficient between annual minimum LAI and annual maximum SPEI from HistDV (1981–2000; left column) and FutDV (2081–2100; right column)

In this study, we employed a drought index (i.e., SPEI) to assess the effects of vegetation dynamics on projected future droughts over West Africa, and impact of vegetation feedbacks on drought characteristics was quantified based on experiments with and without vegetation dynamics. This study suggests that when vegetation dynamics are considered, drought is prolonged and enhanced over the Sahel, and this effect is similar between present and future climates. Specifically, based on the regionally averaged ensembles of SPEIs over the Shael, no drought event is identified in FutSV with the minimum SPEI of − 0.94, whereas 4 drought events across the total period of 42 months are identified by an SPEI as low as − 1.61 in FutDV. For the Guinea Coast, vegetation dynamics has an opposit effect with less drought identified in Fut DV than in FutSV. These results are consistent over four different LBCs. Furthermore, we show that simulated vegetation density (i.e., LAI) was correlated well with annual maximun SPEI with the correlation coefficient of up to 0.91, particularly over the Sahel, a sensitive transition zone where the feedback between land and atmosphere is relatively strong.

While most future drought characterization studies have been conducted without considering the role of interactive vegetation in climate models (e.g., Cook et al. [15]; Huang et al. [29]), this study suggests the importance of biosphere–atmosphere interactions in future drought projections. Dirmeyer et al. ([18]) suggest that land–atmosphere coupling is projected to increase in the future across most of the globe and thus the land surface is likely to play a greater role in amplifying hydrologic extremes, such as severe droughts. Our results suggest that the possible increase in severe droughts caused by climate change could be further enhanced by vegetation feedback.

PET in this study is estimated using the Thornthwaite approach, which considers air temperature as a governing feature of PET. However, there are various other methods to calculate PET. For example, the Penman–Montieth method is more physically realistic but requires a diverse input data set (i.e., humidity, radiation coefficient, and wind speed). Van der Schrier et al. ([64]) calculated the change in the global PDSI using two distinct estimates for PET (e.g., Thornthwaite and Penman-Monteith), and found that PSDI based on two PET estimates are identical in terms of trend, average values, and classifying severe wet or dry periods. Conversely, McVicar et al. ([44]) pointed out that climatic conditions other than temperature may compensate the effect of temperature rise on drought in the future climate. These suggest that our conclusions might vary with different drought indices to some extents.

This study points out the potential of prolonged and enhanced drought events in the Sahel by considering the vegetation dynamics, which has seldom been considered in previous studies. Furthermore, many African countries are expected to experience population growth with some of the fastest relative growth in West African countries, including Niger and Chad (Ahmadalipour et al. [1]). Combined with the high likelihood of prolonged and enhanced drought and population growth, there will likely be an increase in water shortage. This will further exacerbate the future drought risks and will present climate adaptation challenges for managing water needs in the region.

This study was supported by the Basic Science Research Program through the National Research Foundation of Korea, which was funded by the Ministry of Science, ICT & Future Planning (2020R1A2C2007670) and the Internationalization Infra Fund of Yonsei University (2018 Fall semester).

Graph: (PDF 1.50 mb).

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.