The regional impact of Land-Use Land-cover Change (LULCC) over West Africa from an ensemble of global climate models under the auspices of the WAMME2 project
The population of the Sahel region of West Africa has approximately doubled in the past 50 years, and could potentially double again by the middle of this century. This has led to the northward expansion of agricultural areas at the expense of natural savanna, leading to widespread land use -land cover change (LULCC). Because there is strong evidence of significant surface-atmosphere coupling in this region, one of the main goals of the West African Monsoon Modeling and Evaluation project phase II is to provide basic understanding of LULCC on the regional climate, and to evaluate the sensitivity of the seasonal variability of the West African Monsoon to LULCC. The prescribed LULCC is based on the changes from 1950 through 1990, representing a maximum feasible degradation scenario in the past half century. It is applied to 5 state of the art global climate models (GCMs) over a 6-year simulation period. Multiple GCMs are used because the magnitude of the impact of LULCC depends on model-dependent coupling strength between the surface and the overlying atmosphere, the magnitude of the surface biophysical changes, and how the key processes linking the surface with the atmosphere are parameterized within a particular model framework. Land cover maps and surface parameters may vary widely among models; therefore a special effort was made to impose consistent biogeophysical responses of surface parameters to LULCC using a simple experimental setup. The prescribed LULCC corresponds to degraded vegetation conditions, which mainly cause increases in the Bowen ratio and decreases in the surface net radiation, and result in a significant reduction in surface evaporation (upwards of 1 mm day−1 over a large part of the Sahel). This, in turn, mainly leads to less moisture convergence and precipitation over the LULCC zone. The overall impact is a rainfall reduction with every model, which ranges across models from 4 to 25 % averaged over the Sahel, and a southward shift of the rainfall peak in three of the five models which evokes a precipitation dipole pattern which is consistent with the observed pattern for dry climate anomalies over this region. The African Easterly Jet shifts equator-ward, although the strength of this change varies considerably among the models. In most of the models, the main factor causing diabatic cooling of the upper troposphere and enhanced subsidence over the region of LULCC is the reduction of convective heating rates linked to reduced latent heat flux and moisture flux convergence. In broad agreement with previous studies, the impact of degradation on the regional climate is found to vary among the different models, however, the signal is stronger and more consistent between the models here than in previous inter-comparison projects. This is likely related to our emphasis on prioritizing a consistent impact of LULCC on the surface biophysical properties.
KeywordsAfrican monsoon Land use land cover change Land degradation Climate simulations Land surface models Land–atmosphere coupling
The population of the Sahel region of West Africa has approximately doubled in the past 50 years, and could potentially double again by the middle of this century. This increase will put ever more pressure on the already limited water and agricultural resources in the region. In recent years, food production has indeed increased, but this has been mostly due to increases in the surface area cultivated (230 %) rather than actual yield (42 %) (Blein et al. 2008). This increase has led to the expansion of agricultural and pasture areas at the expense of natural savanna and forests (e.g. Leblanc et al. 2008), leading to widespread land use and land cover change (LULCC). These changes can lead to land degradation, and some of the most general causes are overgrazing, continuous cropping, deforestation for firewood, and mismanagement of soil and water resources.
The land surface has been shown to be an important factor in modulating the West African monsoon (WAM). For example, based on observations, the land surface characteristics and processes have been shown to have a significant impact on the inter-annual variability of rainfall in the Sahel region (Nicholson 2013). The importance of surface-atmosphere interactions was one of the main tenets of the recent international African Monsoon Multidisciplinary Analysis (AMMA) project (Redelsperger et al. 2006) and was investigated in several studies (see Taylor et al. 2011, for a summary). This region typically appears as one where the soil moisture feedbacks with the atmosphere are among the strongest over the globe (e.g. Dirmeyer 2011). Using an ensemble of state-of-the-art global climate models (GCMs), the Sahel region has been identified as one of strong soil moisture-atmosphere coupling (Koster et al. 2004). In addition, it has been determined to be the region of the world with the highest impact of biophysical processes on the climate (Xue et al. 2004, 2010b). Indeed many numerical studies have shown the importance of the land surface on modulating the WAM (for a review, see Xue et al. 2012). For example, previous studies have examined the role of changes in the surface albedo (e.g. Charney 1975; Sud and Fennessy 1982; Laval and Picon 1986) and the vegetation (e.g. Xue et al. 1990; Xue 1997; Zheng and Eltahir 1997; Li et al. 2007) on modulating the WAM. All of these studies lead to the general conclusion that reduced vegetation leads to reduced rainfall. Most of these studies were based on sensitivity experiments using single GCMs and land surface models (LSMs), which have model-specific LULC classifications, with idealized and sometimes extreme LULCC scenarios.
There is increasing evidence from numerical studies that anthropogenic LULCC can potentially induce significant variations on the local to regional scale climate (Pielke et al. 2011). However, the IPCC’s Fifth Assessment lacked a comprehensive evaluation of the relative impact of biogeophysical feedbacks of LULCC on regional climate (Mahmood et al. 2014). This is primarily due to over-simplifications and limits to how some key biogeophysical surface processes are represented in the LSM component of GCMs, and how LULC is represented in such models. The recent Land-Use and Climate Identification of robust impacts (LUCID) experiment (Pitman et al. 2009; de Noblet-Ducoudré et al. 2012) examined the biogeophysical impacts of prescribed, global-scale LULCC using an ensemble of coupled GCMs and LSMs. The goal was to identify impacts that were statistically robust, primarily in terms of being detectable, common among the different models, and above the models’ internal variability. LULCC was prescribed based on historical data, with changes based on the time period starting with the beginning of the industrial revolution to present day conditions. LSM models modified their land cover (following their own classification) on the non-crop areas in order to adhere to these changes as much as possible. There turned out to be considerable discrepancy among the models in terms of their response to LULCC, particularly in terms of the Bowen ratio. In addition, LULCC-induced changes in surface sensible heat and latent heat fluxes had opposite signs in some regions among the different models. Reasons for discrepancies are likely related to differences in the land surface parameterizations: for example, how soil moisture is taken up by transpiration, the treatment of turbulence in the surface layer, vadose-zone hydrology and surface runoff production, and the use of dynamic verses prescribed vegetation. The definition of land use classes can also be a source of significant differences. For example, a crop class in one model might be represented as natural grasses in another model: not only can this result in important differences in the values of certain key biogeophysical parameters, it can also cause differences in terms of the imposed seasonal evolution of these parameters. In addition, two models might use the same class for a given vegetation cover, but very different values of the same parameters that characterize the vegetation (such as structural parameters). Different modeling groups have developed special methodologies for changing land cover distributions, which can lead to discrepancies. Another obvious source of differences results from inter-model variations of the simulated coupling strength between the surface and the atmosphere (Koster et al. 2004). In addition to being related to the LSM physics, the coupling strength is also influenced by the physical parameterizations in the different host atmospheric models, notably the parameterizations of planetary boundary layer (PBL, turbulence), convection, clouds, and atmospheric radiation.
One of the main goals of the West African Monsoon Modeling and Evaluation project phase II (WAMMEII) is to provide a basic understanding of LULCC forcing on the regional climate of West Africa (Xue et al. 2016, this issue). The strategy is to apply observational data-based anomaly forcing, i.e., “idealized but realistic”, in GCM and RCM simulations. The prescribed LULCC is based on data from Hurtt et al. (2006) which was used to design a maximum feasible degradation scenario, which will be discussed in Sect. 2.1. In the current study, 5 state-of-the-art GCMs are used in order to study a range of model responses to a common LULCC scenario with a focus on West Africa. In addition, two GCMs use the same LSM (so presumably differences only arise owing to atmospheric effects between these two models), and two models performed two simulations with varying degrees of LULCC. The model intercomparison results for RCMs are reported by Hagos et al. (2014) and Wang et al. (2015). This paper is organized as follows: the methodology for imposing LULCC for multiple GCMs is summarized in Sect. 2, results are presented in Sect. 3, a summary of the main impacts of LULCC on the WAM are summarized in Sect. 4, along with the conclusions of this study.
2 Experimental design
2.1 LULCC methodology
The definition of a control vegetation map is a complicated issue owing to many factors. Due to the errors in satellite data acquisition, data processing, information extraction methodologies, inadequate ancillary training data, as well as the relatively course resolution in current climate models, current satellite-derived vegetation maps have difficulties with respect to adequately present realistic LULC information. This is especially true over West Africa, where the agricultural areas are generally poorly classified. These potential vegetation maps are often based upon remote sensing-produced vegetation maps. This is the current status of terrestrial remote sensing (e.g. Kim et al. 2015), and the experimental design has to take this into account. Thus, in the current study, LULCC is applied to the default current map for each model (which for many models consists in either constant class-based vegetation indices or some sort of climatological annual cycle). The LULCC experiment uses combined crop and pasture fraction changes from Hurtt et al. (2006). This data set was used by the 5th Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC 2014) and was used in the Coupled Model Intercomparison Project (CMIP5) project (Taylor et al. 2012a, b). Note that the dataset used in this study is based on land use change that occurred from 1950 to 1990 which showed a dramatic land degradation in West Africa and became less pronounced (more flat) afterwards (Song 2013). The data has been translated into LULCC for each of the models participating in the present study. Note that the actual changes in land use are not conserved when transformed into model parameter space because each model has their own implementation strategy. Thus, the total LULCC is meant to represent the maximum feasible total amount of degradation resulting from anthropization (i.e. the conversion of natural vegetation, mainly savanna and low trees or shrubs, to cropland or pasture). This is an important distinction from some of the studies mentioned in the previous section (Pitman et al. 2009; de Noblet-Ducoudré et al. 2012), which can potentially include regions where vegetation has returned to a so-called natural state. In addition, another important distinction is that this study prioritizes degradation that is consistent in terms of the biogeophysical response of the surface, as opposed to a consistent land classification change. This is done because two models can impose similar LULCC but produce very different results in terms of the biophysical response, i.e. the values of surface physiographic parameters.
Making consistent LULCC in vegetation maps among models. Also, since significant differences are caused by different map spatial resolutions, a specified LULCC region that is broad enough to be used by even the most coarse resolution models (and has a relatively simple geometrical configuration) was defined. The maximum land class change permitted within a particular grid cell was constrained to be less than or equal to the maximum change estimated from the historical land use data set used in this study over the Sahel for the specified time period.
Proposing the modifications to the land class in the grid points associated with LULCC based on historical land use change information, and determining a set of corresponding surface parameters after careful coordination with each of the different modeling groups.
Maintaining a consistent meridional gradient of the LULCC in terms of the surface vegetation distribution, in an attempt to ensure that the prescribed changes to the surface parameters truly correspond to a consistent degradation in the land maps among the models.
2.2 LULCC implementation
A summary of the WAMME2 models performing the LULCC experiments
GCM (acronym used in Figs.)
LULC (number of classes)
Land surface model
Mechoso et al. (2000)
C. R. Mechoso
2.5° × 2° × L17
Prognostic version of Arakawa and Schubert (1974)
Harshvardhan et al. (1987)
Xue et al. (1991)
Y. Xue, F. De Sales
Moorthi and Suarez (1992)
Chou and Suarez (1994)
Xue et al. (1991)
Zhang and McFarlane (1995)
Iacono et al. (2008)
Oleson et al. (2008)
S., Mahanama, R. Koster
1° × 1.25° × L72
Moorthi and Suarez (1992)
Chou et al. (2001)
Koster et al. (2000)
UKMO HadGEM 2-A
HadGEM2 model development team (2011)
1.25° × 1.875° × L38
Edwards and Slingo (1996)
Essery et al. (2002)
The basic LULCC map was first provided to participants. The land cover fraction change is limited to 30 % in the current study. This threshold was selected in order to avoid any isolated grid points with had anomalously high LULCC that could produce small scale noise in the GCM response since, as seen in Fig. 1, only a few relatively small areas have changes larger than this threshold. This limit was imposed also under the consideration of the selection criteria number 3. After testing the LULCC within the different vegetation maps used by the models, this criteria seems to serve the best to produce a reasonable gradient. The next step was to compare the control and modified LAI and albedo fields across the modeling groups to ensure that the land class changes result in surface changes that are as consistent as possible among the different models. If a change was found to be inconsistent with the other models or not representative of a degradation, further modifications were made to the land cover in order to obtain a similar biogeophysical response. Potentially several iterations with each participating group were used to achieve this goal.
The second LSM strategy can represent multiple LULC within the same grid box using specific sub-grid tile fractions. The HadGEM Met Office Surface Exchange Scheme (MOSES; Essery et al. 2002) uses the International Geosphere–Biosphere Programme (IGBP) global land cover classification (IGBP 1992) with 9 classes. The back ground soil/litter albedo varies in space (Houldcroft et al. 2009) and is independent of land cover class. The vegetation albedo is fixed for each of the 9 classes, and has no temporal variation. The initial LAI values at the start of the 6-year integration period are derived from MODIS satellite products. The MOSES vegetation phenology module was activated, thus the LAI varies in time for each of the tiles (classes present in this zone). The impact of the LULCC on the land cover can be summarized as follows: broadleaf trees decreased by upwards of 30 %, and grasses decrease by approximately half that value, where the existing ratio of C3–C4 grasses was held constant for each grid cell. Shrub lands decrease in the northern Sahel, but increase to the south (mostly in place of decreased forest). The baresoil fraction increases upwards of 30 %. Two LULCC scenarios are analyzed in the current study. For the default LULCC study, the background soil/litter albedo was increased (to be more like values further north: the values were adjusted so that the Sahel average equals 0.35). This resulted in local changes as high as 0.15, but the average increase over the LULCC zone is approximately half of that value). In the sensitivity test (HadGEM-w: see Sect. 3.4), the background albedo was unchanged.
The CAM5 surface scheme (Common Land Model or CLM; Oleson et al. 2008) can represent up to 17 sub-grid tiles within each grid cell (with 16 being plant functional types or PFTs). It was developed for CLM by Lawrence and Chase (2007) and it uses MODIS-based data for trees, and IGBP data for shrubs and grasses, and data from Ramankutty et al. (2008) for crops. For the current study, LAI varies on a daily basis based on linear interpolation of monthly varying LAI, and the annual cycle is fixed for all 6 years. The soil albedo dynamic range (for moisture saturated and totally dry conditions) is specified, and it varies in time as a function of soil moisture. The soil albedo at saturation is entirely determined by soil color which is an independent spatially varying field. The LULCC consisted in changing the forest (PFTs 5 and 7) to grass (PFT 15), and shrub (PFT 11) to baresoil (PFT 1) within the masked zone. The sum of deforestation and desertification areas was limited locally to 30 %. The soil color was modified to obtain albedo changes similar to the other models.
The GMAO surface scheme (Catchment Land Surface Model, CLSM: Koster et al. 2000) uses 6 tiles. The LAI and Greenness fraction (Grn) monthly climatologies are from global satellite observations along with a model-based surface albedo which has been scaled to match the mean seasonal cycle of MODIS satellite observations. Compared to the other GCMs, the GMAO uses a different methodology for assigning parameter values. A vegetation map that evolves in time during the integration (the CTL LULC is from 1952 to 1957) is used. Within the LULCC domain, LAI and albedo vary inter-annually as a result of the time varying tile fractions, while outside of the domain a fixed annual cycle was used. The LULCC for the sensitivity test (GMAO-w; see Sect. 3.4) used the GMAO default long-term integration setup for which only the tile fractions are varied based on data from Hurtt et al. (2006). As it turns out, the sensitivity of surface fluxes to vegetation classes is significantly smaller than that of the prescribed LAI, Grn, and albedo data. Thus, the default LULCC experiment in the current study (referred to as GMAO) required the development of a parameter data set to mimic land use change/degradation over a period of 50 years. In the LULCC experiment, the LAI was approximately halved and the albedo was approximately doubled compared to GMAO-w. This underscores the difficulty in developing a consistent LULCC, especially in terms of the vegetation structural parameters and coverage.
2.3 Experimental protocal
The CTL experiment protocol is described in detail in Xue et al. (2016: this issue), so only a brief summary is provided herein. GCM models are initialized on January 1, 2006, and then run for 6 years. Time varying sea surface temperatures from correspond to a single climatological annual cycle. The first year results are discarded to minimize the potential impact of model spin up. Since the main forcings are the same for all 6 years, the remaining simulations are treated as a 5-member ensemble for each model when doing the results analysis. The atmospheric initial conditions are from the NCEP/DOE (National Center for Environmental Prediction–Department of Energy) Reanalysis II (Kanamitsu et al. 2002). The LULCC experiment uses the same initial conditions and SST forcings, the only difference is in the land cover parameters within the LULCC zone over West Africa.
In this section, “differences” are computed by subtracting mean variables obtained in Exp. CTL from those obtained in Exp. LULCC. The impact of LULCC was found to be largest mainly during the peak monsoon months (here meaning peak Sahelian climatological rainfall, i.e. JAS: July, August, September). In the CTL and LULCC experiments, 61 and 63 %, respectively, of the annual rainfall within the Sahel and Ghana regions (LULCC area shown in Fig. 1) averaged for the 5 GCMs fell during this 3 month period. Therefore, the analysis presented in this study focuses on JAS averages. Note that since the prescribed SSTs are climatological, the last 5 years of the six-year simulations are used as a 5-member ensemble for each of the GCMs and for the multi-model ensemble for computing statistics. A two-tailed student t test is used to test for significance of the local mean values at the 5 and 10 % confidence levels. The significance levels are computed using the pool permutation procedure (PPP), which has been used to inter-compare global multi-model results (e.g. Santer and Wigley 1990). This method accounts for the effects of multiplicity and spatial autocorrelation, and provides estimates of field significance level (p value) for a given number of permutations: 1000 were used to compute statistics in the current study. The impact of LULCC is mainly confined to the region referred to herein as the analysis domain where surface properties were modified, therefore the aforementioned statistics were computed over the domain from 5° to 20° North latitude, and −10°–30° East longitude.
3.1 Impact of LULCC on surface properties
3.2 Impact of LULCC on surface fluxes
Relative difference (%) between the LULCC and CTL experiments of the JAS average rainfall and surface fluxes for each GCM
The correlation and the residual standard error (mm day−1, in parentheses) between changes in each of the two JAS averaged variables, and between rainfall and evapotranspiration within the LULCC zone
LE versus LAI
LE versus albedo
Rainf versus LAI
Rainf versus albedo
Rainf versus LE
3.3 Impact of LULCC on the WAM rainfall
Statistical significance testing results for the JAS average differences between the control and LULCC experiments
3.4 Sensitivity tests
The discussions above show that although the models have substantial differences in their response to the LULCC, they show a general agreement that LULCC which occurred over the prescribed time period has a positive impact on Sahel drought. In previous LULCC multi-model studies, it has been shown that models have difficulties reaching a consensus view of the impact, part of which is due to the difficult task of prescribing LULCC in a consistent manner among different models (Pitman et al. 2009). A similar situation was faced in the early phases of this experiment. In the initial simulations with GMAO and HadGEM, the vegetation type changes based on the classification tables in their respective models only resulted in relatively small changes in one or more of the three key biogeophysical parameters (notably LAI and albedo) which were significantly less compared to the other GCMs, despite adherence to the WAMME guidance. As an example, the initial albedo JAS changes for GMAO and HadGEM are shown in Fig. 4c, e, respectively (compared to the final values for a stronger impact of LULCC shown in Fig. 4d, f, respectively). In this section, results from the simulations for GMAO and HadGEM using the parameter values corresponding to a weak change in biogeophysical parameters are presented and compared to those corresponding to the stronger changes used in the analysis presented in Sects. 3.1–3.3. The main goal of these experiments is to highlight that the application of a common protocol for LULCC can induce big differences in the atmospheric response if the biogeophysical impact is not similar despite the appearance that the models had similar LULCC.
4 Discussion and conclusions
Reducing the LAI increases the Bowen ratio in regions where transpiration and evaporation from intercepted canopy water are occurring. In all of the regions where LAI (and evaporation) decreases (above some relatively low threshold), the rainfall also decreases. This response is common to all the models. Therefore in such regions, resulting decrease in LE is the main cause for reduced rainfall, rather than the reverse.
The increase in albedo reduces the net radiation, thus the energy available for the turbulent surface fluxes are also decreased, but the partitioning of this energy loss is modulated by the LAI change. In models with moderate LAI changes, latent heat flux is reduced during the wet season in regions receiving rainfall. For models with large LAI changes, the reduction in latent heat can exceed the reduction in net radiation caused by the albedo change (owing to large changes in Bowen ratio), thus the sensible heat flux increases. In the dry season or in dry regions (north of the area receiving rainfall), the increase in albedo (reduction in surface net radiation) translates nearly directly into a decrease in sensible heat flux (and there is little to no impact on overall monsoon rainfall).
The model specific simulated WAM location influences the impact of the LULCC. The models with the WAM (defined here as the zone with peak JAS rainfall) located furthest to the north (CAM5) experienced a shift in the overall monsoon position owing to LULCC. This feature is seen as a statistically significant (at the 95 % confidence level) JAS precipitation difference dipole pattern. Two other models (with monsoons located further south) also had a dipole pattern, but rainfall increases were not statistically significant. But the main (and statistically robust) impact in all of the models is a lowering of monsoon rainfall within the LULCC zone: the CAM5 was the only model for which this effect extended outside of this zone (to the north). For the models with a more southerly peak monsoon rainfall (HadGEM and GMAO), there was essentially no southward shift and only a rainfall reduction.
In this WAMME study, the goal is to favor consistent changes in the values of the biogeophysical parameters over changes in a particular model’s LULC, since how vegetation classes and their associated parameter values are defined can vary tremendously between different models. Collaborations were engaged with each modeling group in order to ensure the LULCC experiment not only had a consistent change in the spatial distribution of LULCC and the vegetation types, but also in terms of the vegetation characteristics and parameters, which provide the real forcing at the land surface in the LULCC experiment. But despite these efforts, this remains a challenging task mainly owing to how LULC and the associated biogeophysical parameters are defined in the models.
The impact of LULCC in a single GCM can be quite different from other models in this region owing to several factors. The first is the simulated WAM intensity and location. Most GCMs have continued difficulties simulating the spatial dimensions, temporal evolution, triggering and strength of these systems (e.g.s Hourdin et al. 2010; Roehrig et al. 2013). These results highlight the need for improved multi-model experiments in order to progress on the understanding of LULCC, and for further improvement of the simulation of monsoon systems. In the WAMME I experiment, substantial evaluations of the WAMME models’ performance in simulating the WAM and the surface energy and water balances were made (Xue et al. 2010a; Boone et al. 2010), which provide the base for the WAMME II experiment. The second factor is that the coupling strength is known to be highly variable among models for the same region (Koster et al. 2004), and there is a need to improve the understanding of land/atmosphere interaction using observations (Dirmeyer et al. 2009; Taylor et al. 2012a, b). This region has been highlighted as a region with strong surface-atmosphere coupling (Xue et al. 2010b) since it is in a water-limited transition zone (between arid and wet regions) with a considerable convective rainfall component, and thus it is not surprising that the modeled WAM is sensitive to changes in the surface properties. The third factor is how LULCC is applied (in a relatively consistent manner for multiple models), and how a given LULCC impacts the biogeophysical surface parameters. In broad agreement with the findings of Pitman et al. (2009) and de Noblet-Ducoudré et al. (2012), the effect of LULCC varies considerably among the models, even though an attempt has been made to harmonize the LULCC as much as possible by selecting a relatively simple implementation. LSMs tend to have highly model specific physiographic databases, which have been implicitly tuned to some extent within coupled model systems to the extent that swapping physiographic databases between two different LSMs will likely lead to different climatological features in coupled GCM models. Also, the different approaches to representing the surface features (dominant class, aggregated parameters, explicit tiles for each class present in a GCM grid cell, and the number of classes) can lead to differences in the strength of LULCC for a given grid cell or region. For example we found that the dominant class schemes have a more dramatic LULCC effect than seen in the tile schemes. This underscores the importance of the LSM (and LULCC) implementation.
The present study shows that the LULCC change induces a mostly local effect consistent with the findings of Pitman et al. (2009), as seen in the correspondence between evaporation, net radiation and the precipitation. The first order impact of LULCC was to change the surface fluxes within the LULCC zone, as expected. All of the models had statistically significant changes in surface fluxes over the region of LULCC at the 95 % confidence level. However, the degree to which each of the fluxes was affected varied among the models. For one model, HadGEM, the albedo changes translated directly into an evapotranspiration reduction, which directly impacted (reduced) the rainfall. In two other models (UCLA-GCM and UCLA-AGCM), the albedo reductions again resulted in reduced latent fluxes, but these were co-located with LAI reductions that translated into increased sensible heat flux. In the GMAO model, the biggest albedo changes occurred outside of the main area receiving monsoon rainfall, thus LAI changes were the biggest factor in reducing rainfall. The albedo increase seemed to have relatively little impact when it occurred north of the area receiving rainfall. Three of the models had precipitation dipole patterns related to a decrease in rains within the LULCC zone and a smaller area of increases with a peak located over Cameroon. But the model with the strongest and statistically robust dipole, CAM5, had a large shift of the monsoon rains to the south. This model also has a CTL core monsoon position further north than the other models, and it is the only one that shows a substantial shift in the monsoon position although the impact is still the largest within the LULCC zone. The local link between the rainfall changes and the surface fluxes is more difficult to establish for this model. But note that, as mentioned in Sect. 3.2, the LE is nearly half of that of the other models in the active rain area within the LULCC zone, and this flux is one of the key linkages involved in the coupling of rainfall with the surface at seasonal scale (Xue et al. 2010a; Dirmeyer 2011). Further work needs to be done to determine if this result occurs because this model has a different coupling strength with the atmosphere than the other models. The JAS average rainfall decreased over the LULCC zone (encompassing the Sahel and Ivory coast) by 4–25 % among the models owing to LULCC. The rainfall reductions were almost entirely confined to the LULCC zone except for some rainfall changes which occurred a certain distance downstream of the coast over the eastern Atlantic. This impact was found to dampen out fairly rapidly with increasing distance from the West African coast. In three of the models, the reduction in rainfall was larger than the reduction in evapotranspiration, so that the reduction in moisture convergence was considerably larger than decreases in evapotranspiration. In four of the models, an increased meridional temperature gradient in the lower atmosphere in the LULCC experiment was caused mainly by a decrease in LAI (corresponding with a Bowen ratio increase) generally leading to a the southerly shift of the AEJ (but again, the degree of this shift was also related to the CTL position). Thus, changes in the atmospheric circulation also play an important role (e.g. Xue 1997).
This is essentially a pilot multi-model study for obtaining a better understanding of the effects of LULCC over West Africa. A small number of GCMs, climatological SST forcing resulting in a multi-year ensemble, and a relatively simple methodology for representing LULCC were used in order to focus on elucidating the first order physical mechanisms. This study is an extension of previous LULCC studies in that special attention has been made to have a consistent biogeophysical response (in terms of land surface parameters). In addition, based on these results, it can be inferred that the use of climatological land cover can lead to inconsistencies and errors in GCM studies for West Africa, given the high sensitivity to the surface properties in this region which have a large inter-annual variability, notably the LAI. Inconsistencies can also arise between locations where LULCC is imposed and those of the simulated monsoon (thereby potentially influencing the magnitude of the impact of LULCC). One way to address this issue is to use GCMs that include interactive vegetation schemes within the LSMs (e.g., Wang and Eltahir 2000; Zhang et al. 2015), but such models are still rapidly evolving and they add an additional degree of freedom to the coupled system (it is not clear whether including such feedbacks will dampen or amplify the signal). It is suggested that future multi-model LULCC studies over west Africa include this component at least as a part of sensitivity tests. In terms of the observable impacts of LULCC, there is evidence from satellite data that the Sahel has experienced, on average, a re-greening over the last few decades, and this signal is probably mostly related to a modest recovery in rainfall (Dardel et al. 2014). However, local areas more susceptible to soil degradation (such as those characterized as shallow, sandy soils) have shown a reduction in vegetation cover. In addition, several studies have suggested that the recent expansion of irrigation in Northwest India and Pakistan could be having deleterious effects on monsoon precipitation (Douglas et al. 2009; Saeed et al. 2009; Tuinenberg et al. 2012; Guimberteau et al. 2012; Wei et al. 2013). With more LULCC data available from different sources showing substantial values in past decades (e.g. Hurtt et al. 2011; Kim et al. 2014) in addition to experience gained from previous multi-model studies (Pitman et al. 2009; de Noblet-Ducoudré et al. 2012), LULCC effects within the monsoon system can be more realistically assessed with ensemble runs in parallel with offline LSM simulations with different LULCC scenarios (ranging from weak to strong degradation) in an effort to better quantify the changes in surface parameters required to produce an atmospheric signal (e.g. Xue and Dirmeyer 2015). Finally, it is suggested that future work should be undertaken to evaluate whether the sign and strength of the feedbacks between the surface and the atmosphere simulated by large scale atmospheric models are consistent with observations (Taylor et al. 2012a, b).
This study was supported by the French component of AMMA. Based on French initiative, AMMA was built by an international scientific group and is currently funded by a large number of agencies, especially from France, UK, US and Africa. It has been beneficiary of a major financial contribution from the European Community’s Sixth Framework Research Programme. Detailed information on scientific coordination and funding is available on the AMMA International website http://www.amma-international.org. The authors acknowledge the ESPRI/IPSL database team for hosting the WAMME2 workspace within the framework of the AMMA database, and to K. Ramage, S. Bouffies-Cloche, and L. Fleury for their kind assistance with the WAMME2 database. We wish to acknowledge comments by R. Koster. R. Comer’s contribution was funded by the UK Department for International Development (DFID). The WAMME activity and analysis are supported by U.S. NSF Grants AGS-1115506 and AGS-1419526.
- Blein R, Soulé BG, Dupaigre BF, Yérma B (2008) Agricultural Potential of West Africa. Economic Community of West African States (ECOWAS). Foundation pour l’agriculture et la ruralité dans le monde (FARM). pp 118Google Scholar
- Chou MI, Suarez MJ (1994) An efficient thermal infrared radiation parameterization for use in general circulation models. NASA Tech Mem 104606(3):85Google Scholar
- Chou M-D, Suarez MJ, Liang XZ, Yan M-H (2001) A thermal infrared radiation parameterization for atmospheric studies. Technical report series on global modeling and data assimilation, vol 19, NASA/TM-2001-104606. pp 68Google Scholar
- de Noblet-Ducoudré N, Boisier J, Pitman A, Bonan G, Brovkin V, Cruz F, Delire C, Gayler V, van den Kurk B, Lawrence D, van der Molen M, Müller C, Reick C, Strengers B, Voldoire A (2012) Determining robust impacts of land-use induced land-cover changeson surface climate over North America and Eurasia; Results from the first set of LUCID experiments. J Clim 25:3261–3281. doi:10.1175/JCLI-D-11-00338.1 CrossRefGoogle Scholar
- Douglas EM, Beltrán-Przekurat A, Niyogi D, Pielke RA Sr, Vörösmarty CJ (2009) The impact of agricultural intensification and irrigation on land-atmosphere interactions and Indian monsoon precipitation-a mesoscale modeling perspective. Glob Planet Change 67:117–128. doi:10.1016/j.gloplacha.2008.12.007 CrossRefGoogle Scholar
- Hourdin F, Musat I, Grandpeix J-Y, Polcher J, Guichard F, Favot F, Marquet P, Boone A, Lafore J-P, Redelsperger J-L, Ruti PM, Dell’aquila A, Filiberti MA, Pham M, Doval TL, Traore AK, Gallée H (2010) AMMA-model intercomparison project. Bull Am Meteorol Soc 91:95–104. doi:10.1175/2009BAMS2791.1 CrossRefGoogle Scholar
- Hurtt GC, Frolking S, Fearon MG, Moore B III, Shevliakova E, Malyshev S, Pacala SW, Houghton RA (2006) The underpinnings of land-use history: three centuries of global gridded land-use transitions, wood harvest activity, and resulting secondary lands. Glob Change Biol 12:1208–1229. doi:10.1111/j.1365-2486.2006.01150.x CrossRefGoogle Scholar
- Hurtt GC, Chini LP, Frolking S, Betts R, Feddema J, Fischer G, Fisk JP, Hibbard K, Houghton RA, Janetos A, Jones C, Kindermann G, Kinoshita T, Klein Goldewijk K, Riahi K, Shevliakova E, Smith S, Stehfest E, Thomson A, Thornton P, van Vuuren DP, Wang Y (2011) Harmonization of land-use scenarios for the period 1500–2100: 600 Years of global gridded annual land-use transitions, wood harvest, and resulting secondary lands. Clim Change 109:117–161. doi:10.1007/s10584-011-0153-2 CrossRefGoogle Scholar
- IGBP (1992) Improved global data for land applications. In: Townshend JRG (ed) IGBP Global Change Report No 20. International Geosphere-Biosphere Programme, StockholmGoogle Scholar
- IPCC (2014) Climate change 2014: synthesis report. In: Pachauri RK, Meyer LA (eds) Contribution of working groups I, II and III to the fifth assessment report of the intergovernmental panel on climate change. IPCC, GenevaGoogle Scholar
- Koster RD, Dirmeyer PA, Guo Z, Bonan G, Chan E, Cox P, Gordon CT, Kanae S, Kowalczyk E, Lawrence D, Liu P, Lu CH, Malyshev S, McAvaney B, Mitchell K, Mocko D, Oki T, Oleson K, Pitman A, Sud YC, Taylor CM, Verseghy D, Vasic R, Xue Y, Yamada T (2004) Regions of strong coupling between soil moisture and precipitation. Science 305:1138–1140CrossRefGoogle Scholar
- Kuchler AW (1983) World map of natural vegetation. Goode’s World Atlas, 16th edn. Rand McNally, pp 16–17Google Scholar
- Mahmood R, Pielke RA Sr, Hubbard KG, Niyogi D, Dirmeyer PA, McAlpine C, Carleton AM, Hale R, Gameda S, Beltran-Przekurat A, Baker B, McNider R, Legates DR, Shepherd M, Du J, Blanken PD, Frauenfeld OW, Nair US, Fall S (2014) Land cover changes and their biogeophysical effects on climate. Int J Climatol 34:929–953. doi:10.1002/joc.3736 CrossRefGoogle Scholar
- Mechoso CR, Yu JY, Arakawa A (2000) A coupled GCM pilgrimage: from climate catastrophe to ENSO simulations. General circulation model development: past, present and future. In: Proceedings of a symposium in honor of professor Akio Arakawa, D. A. Randall. Academic Press, USA, pp 539–575Google Scholar
- Pielke RA Sr, Pitman A, Niyogi D, Mahmood R, McAlpine C, Hossain F, Goldewijk KK, Nair U, Betts R, Fall S, Reichstein M, Kabat P, de Noblet N (2011) Land use/land cover changes and climate: modeling analysis and observational evidence. WIREs Clim Change 2:828–850. doi:10.1002/wcc.144 CrossRefGoogle Scholar
- Pitman AJ, de Noblet-Ducoudré N, Cruz FT, Davin EL, Bonan GB, Brovkin V, Claussen M, Delire C, Ganzeveld L, Gayler V et al (2009) Uncertainties in climate responses to past land cover change: first results from the LUCID intercomparison study. Geophys Res Lett 36:L14814. doi:10.1029/2009GL039076 CrossRefGoogle Scholar
- Song G (2013) Global and Sahel regional biophysical processes, vegetation dynamics, and climate interactions. PhD Dissertation. University of California, Los Angeles, p 183Google Scholar
- Taylor CM, Parker DJ, Kalthoff N, Gaertner MA, Philippon N, Bastin S, Harris PP, Boone A, Guichard F, Flamant C, Grandpeix J-Y, Cerlini P, Baldi M, Descroix L, Douville H, Polcher J, Agusti-Panareda A (2011) New perspectives on land-atmosphere feedbacks from the African monsoon multidisciplinary analysis (AMMA). Atmos Sci Lett 12:38–44. doi:10.1002/asl.336 CrossRefGoogle Scholar
- Xue Y, Dirmeyer PA (2015) Land-atmosphere interactions in monsoon regimes and future prospects for enhancing prediction. CLIVAR Exchanges 19(66):28–33Google Scholar
- Xue Y, De Sales F, Lau KMW, Boone A, Feng J, Dirmeyer P, Guo Z, Kim KM, Kitoh A, Kumar V, Poccard-Leclercq I, Mahowald N, Moufouma-Okia W, Pegion P, Rowell D, Schubert SD, Sealy A, Thiaw WM, Vintzileos A, Williams S, Wu MLC (2010a) Intercomparison and analyses of the climatology of the West African monsoon in the West African Monsoon Modeling and Evaluation project (WAMME) First Model Intercomparison Experiment. Clim Dyn 35:3–27. doi:10.1007/s00382-010-0778-2 CrossRefGoogle Scholar
- Xue Y, De Sales F, Lau WK-M, Boone A, Kim K-M, Mechoso CR, Wang G, Kucharski F, Schiro K, Hosaka M, Li S, Druyan LM, Seidou Sanda I, Thiaw W, Zeng N, Comer RE, Lim Y-K, Mahanama S, Song G, Gu Y, Hagos SM, Chin M, Schubert S, Dirmeyer P, Leung LR, Kalnay E, Kitoh A, Lu C-H, Mahowald NM, Zhang Z (2016) West African monsoon decadal variability and surface-related forcings: second West African Monsoon Modeling and Evaluation project experiment (WAMME II). Clim Dyn. doi:10.1007/s00382-016-3224-2 Google Scholar