Northward Shifts of the Sahara Desert in Response to Twenty-First-Century Climate Change

Chuyin Tian aFaculty of Engineering and Applied Science, University of Regina, Regina, Saskatchewan, Canada

Search for other papers by Chuyin Tian in
Current site
Google Scholar
PubMed
Close
,
Guohe Huang aFaculty of Engineering and Applied Science, University of Regina, Regina, Saskatchewan, Canada

Search for other papers by Guohe Huang in
Current site
Google Scholar
PubMed
Close
https://orcid.org/0000-0003-4974-3019
,
Chen Lu aFaculty of Engineering and Applied Science, University of Regina, Regina, Saskatchewan, Canada

Search for other papers by Chen Lu in
Current site
Google Scholar
PubMed
Close
,
Tangnyu Song aFaculty of Engineering and Applied Science, University of Regina, Regina, Saskatchewan, Canada

Search for other papers by Tangnyu Song in
Current site
Google Scholar
PubMed
Close
,
Yinghui Wu aFaculty of Engineering and Applied Science, University of Regina, Regina, Saskatchewan, Canada

Search for other papers by Yinghui Wu in
Current site
Google Scholar
PubMed
Close
, and
Ruixin Duan bState Key Joint Laboratory of Environmental Simulation and Pollution Control, School of Environment, Beijing Normal University, Beijing, China

Search for other papers by Ruixin Duan in
Current site
Google Scholar
PubMed
Close
Free access

Abstract

The spatial extent of the Sahara (the largest nonpolar desert) has significant impacts on the livelihood of people residing in its surrounding areas. Despite the fact that climate change would foreseeably impact the location and size of the desert, its future responses (i.e., advance or retreat) are rarely explored in previous studies. Here, through the development of an ensemble Bayesian discriminant analysis approach, we use 10 of the latest high-resolution GCM (global climate model) simulations to document robust annual and seasonal responses of the Sahara Desert to twenty-first-century climate change, with the consideration of modeling uncertainties. We find northward shifts of the Sahara/Sahel and eastern expansion of the nondesert zone under both SSP2–4.5 and SSP5–8.5 scenarios, the former more pronounced in the wet season and the latter in the dry season. Countries located near the Mediterranean may thus experience higher risks of drought, while the projected retreat of the Saharan southern boundary will be beneficial to the local water availability of proximal countries.

Significance Statement

Given that sub-Saharan Africa is one of the most vulnerable regions to climate change, the Sahara’s expansion would bring unexpected health risks to billions of people. It is thus vital to understand its robust response to global warming. However, previous studies are merely focused on using a simple precipitation threshold as the definition criterion to estimate the varying size of the Sahara Desert. In addition, significant uncertainty in precipitation projections also limits relevant investigations of the Sahara’s future responses. Here, by developing an ensemble Bayesian discriminant analysis approach, we could provide an objective basis for desert identification under large intermodel uncertainty. Further, we find significant northward shifts of both the Sahara and the Sahel, which may induce higher risks of drought over the northwest of North Africa.

Lu’s current affiliation: Earth System Physics Section, The Abdus Salam International Centre for Theoretical Physics, Trieste, Italy.

Duan’s current affiliation: National Institute of Natural Hazards, Ministry of Emergency Management of China, Beijing, China.

© 2023 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Guohe Huang, huangg@uregina.ca

Abstract

The spatial extent of the Sahara (the largest nonpolar desert) has significant impacts on the livelihood of people residing in its surrounding areas. Despite the fact that climate change would foreseeably impact the location and size of the desert, its future responses (i.e., advance or retreat) are rarely explored in previous studies. Here, through the development of an ensemble Bayesian discriminant analysis approach, we use 10 of the latest high-resolution GCM (global climate model) simulations to document robust annual and seasonal responses of the Sahara Desert to twenty-first-century climate change, with the consideration of modeling uncertainties. We find northward shifts of the Sahara/Sahel and eastern expansion of the nondesert zone under both SSP2–4.5 and SSP5–8.5 scenarios, the former more pronounced in the wet season and the latter in the dry season. Countries located near the Mediterranean may thus experience higher risks of drought, while the projected retreat of the Saharan southern boundary will be beneficial to the local water availability of proximal countries.

Significance Statement

Given that sub-Saharan Africa is one of the most vulnerable regions to climate change, the Sahara’s expansion would bring unexpected health risks to billions of people. It is thus vital to understand its robust response to global warming. However, previous studies are merely focused on using a simple precipitation threshold as the definition criterion to estimate the varying size of the Sahara Desert. In addition, significant uncertainty in precipitation projections also limits relevant investigations of the Sahara’s future responses. Here, by developing an ensemble Bayesian discriminant analysis approach, we could provide an objective basis for desert identification under large intermodel uncertainty. Further, we find significant northward shifts of both the Sahara and the Sahel, which may induce higher risks of drought over the northwest of North Africa.

Lu’s current affiliation: Earth System Physics Section, The Abdus Salam International Centre for Theoretical Physics, Trieste, Italy.

Duan’s current affiliation: National Institute of Natural Hazards, Ministry of Emergency Management of China, Beijing, China.

© 2023 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Guohe Huang, huangg@uregina.ca

1. Introduction

Africa is highly vulnerable to climate change, mainly on account of its strong economic dependency on climate-related activities and products with low adaptive capacity (Boko et al. 2007; Liu et al. 2008; Niang et al. 2014). The Sahara of Africa, the largest nonpolar desert in the world, is likewise an attractive study area under climate change. The effects of global warming on the Sahara are of great importance to the proximal countries owing to its profound impacts on local livelihood and socioeconomic development (Viatte et al. 2009; Battisti and Naylor 2009; Nicholson et al. 2012). Moreover, as the Sahara is the world’s largest dust source, its windblown dust plays a vital role in Earth’s climate system, which can alter the atmospheric radiation budget, facilitate cloud formation, and affect the ocean carbon cycle (van der Does et al. 2020; Skonieczny et al. 2019). The long-term response of the Sahara Desert to climate change, expansion, or contraction, thereby receives extensive attention in current climate science.

Many studies have suggested that the wet regions are getting wetter and the dry ones drier under the global warming trend (Chou et al. 2013; Sun et al. 2016; Liu et al. 2022). This may result in severe impacts on the Sahara Desert: the east–west-beltlike shape of the Sahel may become wider with a sharper north–south rainfall gradient. As such, the Sahel’s precipitation variability has been investigated in several studies. Under twentieth-century climate change, it has been detected to be associated with the sea surface temperature (SST) anomalies in the tropical Atlantic and Indian Oceans and the tropical Pacific (Folland et al. 1986; Rowell 2001). Since the 1990s, the summer precipitation of the Sahel displays a significant increase (Li et al. 2012). The amplified warming over the Sahara (Cook and Vizy 2015), the enhanced meridional energy gradient (Dong and Sutton 2015), and the increased anthropogenic regional SST (Maidment et al. 2015) have been demonstrated to aid in the Sahel rainfall recovery. Nevertheless, whether the Sahel rainfall will continue to increase when affected by anthropogenic greenhouse gas is still not very clear yet.

The Sahara’s expansion may affect water availability in countries adjacent to its boundary, exacerbate the vulnerability of their agriculture systems, and pose threats to food security (Niang et al. 2014). Thus, several studies have focused on estimating the varying size of the Sahara Desert (Tucker and Nicholson 1999; Tucker et al. 1991; Thomas and Nigam 2018; Liu and Xue 2020). Due to the close relationship between the vegetation index and precipitation in the Sahara, and the unavailability of a satellite vegetation index (when precipitation is below ∼150–200 mm yr−1), 200 mm yr−1 precipitation isoline has been determined as the boundary between the Sahara Desert and the Sahel. This precipitation threshold is then used to explore the Saharan interannual variations from the 1980s to 1990s (Tucker and Nicholson 1999; Tucker et al. 1991). Similarly, based on two thresholds (100 and 150 mm yr−1), the century-long (1902–2013) seasonal variations have also been investigated, which reveals southward expansion of the Sahara in summer and evident retreat of the northern edge in winter (Thomas and Nigam 2018). With similar methods, future annual variations of the Sahara Desert under climate change have been projected based on two simulations with/without considering the dynamic vegetation process (Liu and Xue 2020).

In previous studies, a desert was generally defined with respect to a threshold of multiyear or seasonal average precipitation. However, the basis for determining such a threshold was generally subjective. Moreover, there were high uncertainties regarding the precipitation projections from multiple GCMs, which would yield significant disagreement over desert identification. Previous research that relied on a single climate model cannot reflect such uncertainties and therefore fails to further address the subsequent disparity of identification results. Thus, an effective multi-GCM ensemble approach is necessary to create robust projections for the advance/retreat of the Sahara Desert.

The focus of this study is to explore the future responses (advance/retreat) of the Sahara Desert to twenty-first-century climate change. Such an effort will be facilitated through the development of an ensemble Bayesian discriminant analysis (EBDA) approach based on 80-yr precipitation projections from an ensemble of CMIP6 models. In detail, we first initiate a definition of the desert by establishing representative vegetation-based zones based on the grid information of the normalized difference vegetation index (NDVI). Then, we examine the probabilistic distribution of precipitation for each zone and create a statistical relationship that connects the vegetation cover and precipitation level. From such a relationship, the probability that a grid point, associated with dedicated precipitation characteristics, belongs to each vegetation-based zone can thus be inferred. Thereafter, we examine the performance of the latest high-resolution (approximately 100 km) CMIP6 models (Eyring et al. 2016) and obtain a subset of the 10 best-performing ones. Based on the precipitation outputs of these models, we then project seasonal and annual Saharan responses to climate change under two shared socioeconomic pathways (SPPs; i.e., SSP2–4.5, and SSP5–8.5). The models evidently disagree among themselves in such responses; through exploiting the probabilistic features of the inference results, we are able to project ensemble responses as endorsed by statistical tests. We also investigate the statistical significance in terms of the effects from various uncertainty sources (i.e., climate models, forcing scenarios, and their interaction) on the projected responses and present their spatial variability through map visualization. Furthermore, we evaluate the plausible consistency of the Saharan responses with future changes in the meteorological systems.

2. Data and methods

a. Overview of the study area

North Africa is one of the most vulnerable regions due to its high exposure and low adaptive capacity to climate change. Not surprisingly, climate change in North Africa is bringing huge challenges to local water resource availability. Several studies indicate a future decrease in water abundance under climate change (Droogers et al. 2012; Notter et al. 2013; Huang et al. 2022). The food production system is also among the world’s most vulnerable since it highly relies on rainfed crop production, with persistent poverty limiting the capacity to adapt (Boko et al. 2007; Schuster-Wallace et al. 2022). Although progress has been made to improve safe water and food coverage, countries in sub-Saharan Africa still show the lowest coverage, highlighting the high health risks under climate change (Battisti and Naylor 2009; Burke et al. 2009; Pan et al. 2022). The expansion of the desert implies that more regions face severe drought. The variations of the Saharan spatial extent could have profound impacts on North African livelihoods and economic stability, as well as social development (Viatte et al. 2009; Battisti and Naylor 2009; Nicholson et al. 2012). Some countries (especially over the Sahel) may experience significant seasonal advances of the Sahara due to fluctuations in regional climate. This would exacerbate these regions’ vulnerabilities to climate change, posing challenges to regional agriculture and water resource planning. Therefore, the study focuses on the future seasonal variations of the Sahara Desert in both dry seasons (December–February and March–May) and wet seasons (remaining months). The annual variations will also be complemented based on annual mean precipitation over North Africa.

b. Data

For the determination of vegetation-based zones, the normalized difference vegetation index observed from the Advanced Very High Resolution Radiometer satellite (AVHRR-NDVI), with a grid resolution of 0.05°, is used for supporting the clustering analysis via the self-organizing map (SOM). The monthly AVHRR-NDVI during the latest period (from 2015 to 2019) is chosen since it reflects recent vegetation distribution in North Africa; it is downloaded from https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc:C00813. The observed information of monthly total precipitation (2015–19, a grid resolution of 1°) is acquired from the Global Precipitation Climatology Centre (GPCC) dataset (https://opendata.dwd.de/climate_environment/GPCC/html/gpcc_monitoring_v6_doi_download.html). It covers more than 200 years involving more than 85 000 observation stations worldwide (Becker et al. 2013; Schneider et al. 2011). Despite the limited stations over the Sahara Desert, the availability of monthly precipitation data (i.e., total number of stations) in GPCC ranks the highest during the period 1901–2011 compared with those from other large-scale climate datasets [Fig. 2 in Schneider et al. (2014)]. With a wide range of countries and regional suppliers providing gauged data, the GPCC data product has been used as the quality-controlled observational dataset in many previous studies (Hu et al. 2018).

Furthermore, 14 recently released CMIP6 models (Table 1, Eyring et al. 2016) with 100-km resolution are selected as candidates to provide historical simulations and future projections of precipitation from 2015 to 2099. The historical runs for the period of 1995–2014 are considered as the baseline period to assess the model’s simulation performance and support the comparisons between historical and projected responses of the Sahara Desert. Projections under SSP2–4.5 and SSP5–8.5 are collected for the Bayesian discriminant analysis (BDA). Two 20-yr periods (2040–59 and 2080–99) are used to analyze the future advance/retreat of the Sahara Desert. For the multi-GCM ensemble, all precipitation information is interpolated to the 1° × 1° grid of the GPCC. As shown in Fig. S1 in the online supplemental material, despite the uncertainty that exists among the 14 ensemble candidates, the majority of them imply increasing rainfall in Northern Africa until the end of this century. Will the rising precipitation facilitate the Sahara Desert’s retreat both annually and seasonally? This question is the focus of our study, which will be addressed with the proposed method.

Table 1

Performances of 14 CMIP6 models based on period averages of annual precipitation.

Table 1

To examine the potential drivers of the Saharan responses to climate change, monthly geopotential height, relative humidity (RH), specific humidity, and u and υ wind components at 925 hPa are obtained from the 14 CMIP6 models (including historical simulations over 1995–2014 and projections over 2080–99). Other variables including surface temperature, total cloud cover percentage, and total outgoing longwave radiation (OLR) are also acquired for the same periods to support the mechanism analysis in section 3e.

c. Development of ensemble Bayesian discriminant analysis approach

Previous studies generally rely on relatively simple approaches (i.e., precipitation thresholds) to estimate the variations in the spatial extent of the Sahara Desert. On the other hand, significant uncertainties exist in climate projections, leading to further uncertainties in the projected desert extent. Therefore, EBDA is developed to address the above challenges, with the detailed schematic illustration (based on period averages of annual precipitation) shown in Fig. 1.

Fig. 1.
Fig. 1.

Schematic illustration for the developed EBDA.

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

1) Establishment of representative vegetation-based zones

SOM (Kohonen 1982) is a topologically sensitive clustering technique. Owing to its good capability in handling multidimensional inputs (Bigdeli et al. 2022; Bação et al. 2005; Bagan et al. 2005), SOM has been extensively used for analyzing multidimensional remote sensing information (e.g., NDVI and land surface temperature) in previous studies (Lasaponara et al. 2020; Lamjiak et al. 2016; Pérez-Hoyos et al. 2014). It generates clusters based on the Euclidean distance of multidimensional input data.

In this study, seasonally (wet and dry) averaged AVHRR-NDVI data for each year (2015–19) are input to SOM for clustering vegetation-based zones (Fig. S2), which can reflect the interannual variability of NDVI. In detail, based on the information of seasonal mean precipitation, yearly averaged AVHRR-NDVI data at each grid in each season (dry or wet) of 2015–19 are used for training SOM to generate two clusters (desert and nondesert zones). The resulting differences can help identify regions that are significantly affected by seasonal alternations. To consider the effects of both dry and wet seasons, the time series of both dry- and wet-season averages for each of the years 2015–19 (i.e., two seasons for 5 years, leading to 10 dimensions) are analyzed via SOM to classify transition, desert, and nondesert zones. Detailed descriptions of SOM are provided in section 2 of the supplemental material.

2) Projection of the seasonal and annual responses of the Sahara Desert to climate change

BDA has been widely used to reflect statistical features of objects from several predefined groups and then assign new sample(s) to respective group(s); it involves the analysis of posterior probability based on Bayes’ principles (Hastie et al. 2009; Srivastava et al. 2007; Li and Xie 2021). Earlier studies have established that precipitation in North Africa displays a significant correlation with NDVI (Tucker et al. 1991; Tucker and Nicholson 1999; Herrmann et al. 2005). The annually and seasonally averaged precipitation is therefore selected as the discriminant variable for each preclustered vegetation-based zone. The seasonal and annual responses of the Sahara Desert can then be projected through the BDA. The detailed steps are summarized as follows:

  1. Based on the preclustered vegetation-based zones (i.e., desert, transition, and nondesert zones, where the transition zone is identified based on the information of annually averaged NDVI), the period averages of annual and seasonal precipitation at grid level are extracted from both GPCC dataset and CMIP6 models.

  2. For each zone type, the mixed Gaussian distribution [Eqs. (1) and (2)] (Titterington et al. 1985; Ansell and Valle 2021) is used to fit the observed data from GPCC, at annual and seasonal scales. The probability distribution of precipitation for each preclustered zone is then established; it reflects statistical heterogeneity among the two (or three) vegetation-based zones:
    fi(x|θ)=iϕifi(x|μi,σi2) and
    fi(x|μi,σi2)=12πσiexp[(xμi)22σi2],
    where φi, μi, and σi2 denote the weight, mean, and variance for the ith term in the mixed Gaussian distribution, respectively. Then the Anderson–Darling test (Razali and Wah 2011) is undertaken to examine the goodness-of-fit of the generated distributions (p value > 0.05 indicates good-fitting performance).
  3. Based on the obtained distributions, for a given precipitation level [xj, obtained from CMIP6 models in step (i)], the posterior probability p(πk|xj) for grid j to belong to zone k (k = 2 at the seasonal scale and k = 3 at the annual one) can be estimated as follows:
    p(πk|xj)=p(xj|πk)p(πk)p(xj)=p(xj|πk)p(πk)k[p(xj|πk)p(πk)],
    where p(πk) represents the prior probability of zone k [p(πk) = number of grids in zone k divided by the total number of grids in all zones]; and p(xj|πk) denotes the probability density for grid j based on the generated probability distributions of zone k in step (ii) [Eq. (4)]:
    p(x|πk)=fi(x|θ) and
    p(x|πm)p(πm)>p(x|πn)p(πn),formn.
    With the obtained p(πk|x), each grid can be classified to a zone type [i.e., the type with the highest posterior probability level; Eq. (5) (see Fig. 1)].

3) Projection of ensemble responses of the Sahara Desert to climate change

A number of uncertainties exist in precipitation projections from multiple CMIP6 models, leading to considerable discrepancies in the posterior probabilities generated from BDA. To address such uncertainties, the t test (Student 1908) is utilized to compare the posterior probabilities in each vegetation-based zone at the seasonal scale; one-way ANOVA (Girden 1992) and post hoc analysis (Rodger 1975) are undertaken for comparisons at the annual scale. The ensemble responses (advance/retreat) of the Sahara Desert can then be projected.

Specifically, after BDA is undertaken for each CMIP6 model, three or two sets of probabilities (depending on the number of preclustered vegetation-based zones) are obtained at each grid; each set includes multiple posterior probabilities corresponding to the ensemble modeling members (see Fig. 1). At the seasonal scale, two sets of probabilities are obtained. Then the t test (Student 1908) is undertaken to examine whether the means for these two sets are statistically different from each other. For three sets of probabilities at the annual scale, using the t test would increase the chance of type I errors (Kucuk et al. 2016). Thus, one-way ANOVA is applied to test whether the means of three probability sets are statistically equal. If the means are not all equal, Tukey’s honestly significant difference (HSD) test (Abdi and Williams 2010) for post hoc analysis is used to test the statistical equality of pairwise mean:
H0:μ1=μ2H1:μ1μ2ifμ1>μ2>μ3(ifpvalue>0.05,acceptthenullhypothesis),
where μ1, μ2, μ3 represent the means of posterior probabilities in the three vegetation-based zones (μ1 > μ2 > μ3). Since each grid will be assigned to the zone type with the largest probability, whether the largest probability is statistically different from the second largest one should be examined. Such an examination (i.e., Tukey’s HSD test) is based on the following statistics:
Tα=qα(a,f)MSEn and
q=y¯Ly¯SMSE/n,
where q represents the distribution of the studentized range statistic; a is the number of sets; f is the degrees of freedom associated with within-group mean square MSE; n is the number of probability levels in each set; y¯Ly¯S are the larger and smaller of the two means being compared, respectively. If a statistically significant difference exists, then the grid can be assigned to the corresponding zone type that displays the highest mean. If not, then this grid would be determined as zone type 4 (statistically indistinguishable, as shown in Fig. 1). As such, the spatial extent of the Sahara Desert can be projected based on the CMIP6 ensemble.

3. Results

a. Model validation and selection

It is essential to examine the performance of currently available CMIP6 models in the historical period before using their precipitation projections. In this study, historical validation of 14 CMIP6 models with 100-km resolution is conducted based on their capability to simulate the regional climatology of North Africa. Fig. S3 presents the spatial distribution of annual mean precipitation for observations (GPCC) and the 14 CMIP6 models. The observed annual means vary from 0 to 150 mm month−1 across North Africa. The lowest annual mean is recorded over the sub-Saharan region, while the highest one occurs over the southern region. Most models can accurately reproduce the annual climatology of the study region. Some individual models (i.e., FGOALS-f3, GFDL-ESM4, INM-CM4, and BCC-CMSM2-MR), however, show evident over- or underestimations in precipitation over some regions. Further, spatial patterns of biases for CMIP6 models against GPCC observations are displayed in Fig. 2 (biases for dry- and wet-season averages are shown in Figs. S4 and S5). FGOALS-f3 shows an underestimation (around 50–100 mm month−1) in the southern parts of the study domain. In contrast, GFDL-ESM4 and BCC-CMSM2-MR demonstrate overestimations (reaching 150 mm month−1 in some regions) over a vast area in the southwest and south of the study area, respectively. INM-CM4 depicts moderate under- and overestimation over western and eastern regions. CESM2-WACCM and NorESM2-MM outperform other models and reasonably capture the observed spatial pattern.

Fig. 2.
Fig. 2.

Spatial patterns of biases for annual mean precipitation (generated from monthly total precipitation over the period 1990–2014) of CMIP6 outputs against observed records (the GPCC).

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

Moreover, the temporal correlation (based on the monthly time series of 1990–2014) between the GCM outputs and observations at the grid level is presented in Fig. S6. Most of the models exhibit acceptable performance over southern parts, having their correlation coefficients (r) lie within the range of 0.5–0.75. A few models (e.g., CMCC-CM2 and AWI-CM) exemplify better performance with r being greater than 0.75 over a large area, which implies appreciable consistency with the observations. Conversely, models that display unsatisfactory performance include FGOALS-f3 and BCC-CSM2-MR, with many grids showing a lower correlation (<0.5). As for the sub-Saharan region, most of the area has relatively limited precipitation during the whole year, which makes it difficult to accurately simulate temporal correlation over this region. To further assess the performance of 14 CMIP6 models, spatial correlation coefficient, normalized root-mean-square error (nRMSE), determination coefficient (R2), and interannual variability skill score (IVS) (Chen et al. 2011) are used as the evaluation metrics (Table 1). In detail, the yearly mean of monthly total precipitation is acquired in each year of 1990–2014 for each grid over North Africa. The obtained multiyear and multigrid means can then be spatially averaged to generate the desired IVS. The lower the IVS is, the better performance the model shows for simulating interannual variability. As listed in the table, the relatively small IVS values show reasonable reproduction of local interannual variability by all models (especially for FIO-ESM-2-0 and INM-CM5-0). The remaining indices (i.e., spatial r, nRMSE, and R2) are estimated based on the period averages of annual, dry-season, and wet-season precipitation at the grid level, which helps identify good-performing CMIP6 models. Consistent with previous findings, for annual mean precipitation, FGOALS-f3, GFDL-ESM4, INM-CM4, and BCC-CMSM2-MR demonstrate evident weakness (low R2 and high nRMSE) in reproducing the observed spatial variability. In comparison, CESM2-WACCM and NorESM2-MM consistently show the best performance, with R2 being greater than 0.85 for annual, as well as dry- and wet-season means.

Precipitation levels, especially the temporal average at the grid level, are important in this study as they are direct inputs of the developed EBDA. Models showing large systematic biases in the temporal mean of precipitation should be excluded. Previous research (Babaousmail et al. 2021) has demonstrated that R2 being around 0.6 implies a reasonable performance of CMIP6 models over North Africa (as confined with 19°–37°N, 20°W–38°E in their study). However, apart from BCC-CMSM2-MR, which has R2 < 0.6, significant under- and/or overestimation can also be found in FGOALS-f3, GFDL-ESM4, and INM-CM4, all with R2 < 0.8 for annual mean precipitation. Therefore, these four models are excluded from the subsequent analysis. A subset of 10 best-performing CMIP6 models is generated, which also display reliable performance in dry and wet seasons (Tables S1 and S2), all with values of R2 > 0.65. The projections of the Saharan responses to climate change are then explored based on this GCM subset owing to their good performance in capturing the observed patterns.

b. Determination of vegetation-based zones

Based on annually or seasonally averaged information of AVHRR-NDVI, three (i.e., desert, transition, and nondesert zones, Fig. S2a) or two (i.e., desert and nondesert zones, Figs. S2b,c) vegetation-based zone types are obtained through SOM. The spatial average of NDVI corresponding to each type is calculated to verify the reliability of clustering results. As shown in Table 2, NDVI means are less than 0.15 for the desert zone and greater than 0.25 for the nondesert zone. The annual mean for the transition zone is between 0.10 and 0.20. These results are consistent with previous studies (Sharma et al. 2019; Anyamba et al. 2014; Kim et al. 2017).

Table 2

Spatial-average NDVI and corresponding misclassification rate obtained from the developed EBDA.

Table 2

With the determined vegetation-based zones, precipitation within the corresponding spatial extent can be extracted. We then examine the probability distribution of precipitation for each zone type; these distributions are used as the basis for the subsequent BDA. Specifically, annual and seasonal (dry and wet seasons) means of gridded observational data from the GPCC in previously defined zones are collected, respectively, for distribution fitting. The empirical and fitted distribution of precipitation for desert, nondesert, and/or transition zones are shown in Figs. 3 and 4.

Fig. 3.
Fig. 3.

Empirical and fitted probabilistic and cumulative distribution functions for precipitation in (a) desert, (b) transition, and (c) nondesert zones. The number of sample points is reduced for better visualization. Gridded annual-mean precipitation levels are extracted from the GPCC (for desert, transition, and nondesert zones).

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

Fig. 4.
Fig. 4.

As in Fig. 3, but for gridded seasonal-mean precipitation levels extracted from the GPCC (for desert and nondesert zones).

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

The Anderson–Darling test (Lu et al. 2017) is applied to assess the goodness-of-fit of the fitting. The results indicate that the empirical cumulative density functions (ecdfs) are well fitted by the mixed Gaussian distribution (log transformation of precipitation data is conducted for a better fitting). As shown in Fig. S7, the distributions of desert and nondesert zones in the dry season are partially overlapped around 4–6, which results in a slightly higher error rate (11%) of BDA than that for annual mean precipitation (9%). Precipitation levels (after log transformation) over North Africa in the dry season are mostly distributed within a narrow range of −4 to 8, regardless of different zone types; this may be the main reason for the slightly high error rate. By contrast, precipitation shows evident differences between the desert zone and the nondesert one during the wet season. Hence, their probability density functions (pdfs) are apparently distinguishable, leading to the lowest error rate (8%) of BDA (Table 2).

c. Uncertainties in the projected responses of the Sahara Desert to climate change

Discrepancies exist in precipitation projections among GCMs, which will lead to uncertainties in the projections of the Sahara’s responses to climate change. To reflect such uncertainties, the results from BDA are investigated in this study. In detail, the projections of vegetation zone types for each individual GCM are obtained based on its annual/seasonal mean precipitation. Subsequently, the frequency of each zone type for each grid (i.e., the number of models that support this grid being classified as each type divided by the total number of CMIP6 models) over two periods (i.e., 2040–59 and 2080–99) is estimated to represent uncertainties in the projected responses of the Sahara Desert to climate change.

For both scenarios (SSP2–4.5 and SSP5–8.5), as shown in Fig. 5 (and Figs. S8 and S9), despite significant uncertainties (low level of frequency) in the boundaries of three zones, the entire Sahara Desert displays a tendency to move northward over time. That is, more grids display decreased frequencies in the desert type on the southern border and increased ones over the northwestern boundary. A clearer pattern is shown under SSP5–8.5, with fewer grids being classified as desert over the southern border. As expected, for the frequency of the transition zone, substantial intermodel spreads around the Sahel can be found, as shown in Fig. S8, with few grids having values exceeding 0.75. Moreover, under SSP5–8.5, uncertainties over border regions are reported to increase over time, especially over those between the transition zone and the nondesert one. For the frequency of the nondesert zone (Fig. S9), different from the slight tendency under SSP2–4.5, the western and middle parts display consistent increasing trends under SSP5–8.5; namely, the number of grids whose frequency = 1.0 decreases significantly throughout the twenty-first century. In addition, the 10 CMIP6 models show the least agreement over the eastern part of North Africa (e.g., the Horn of Africa). Such a pattern may impact further projection of the Sahara’s ensemble responses.

Fig. 5.
Fig. 5.

The spatial distribution of frequency for the desert under (left) SSP2–4.5 and (right) SSP5–8.5. At each grid, based on annual means of precipitation, the frequency is estimated by the number of models that support certain grids being classified as desert divided by the total number of CMIP6 models.

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

BDA results based on seasonal mean precipitation tell quite different stories. For the results based on dry-season means (Figs. S10 and S11), the number of grids that are exclusively classified as desert (i.e., frequency = 1) would increase from 120 (in 2040–59) to 128 (in 2080–99) over the northwest of North Africa (30°–36.5°N, 15°W–15°E); this demonstrates a slightly increased tendency of desertification under SSP5–8.5. More models support that grids in these regions should be classified as desert, suggesting that the northwestern boundary of the Sahara Desert is more likely to migrate northward in response to climate change. Similar patterns can be found under SSP2–4.5. Furthermore, the frequency of desert in the eastern region declines under both scenarios (more apparent in SSP5–8.5), suggesting that a number of grid cells here are more likely to undergo evolution from the desert to the nondesert zone. Such a pattern is also shown in the wet season (Fig. S12). However, different from negligible variations in dry-season results, evident increases in uncertainty can be found over the desert-nondesert boundary from the mid- to late twenty-first century (Fig. S13). On the other hand, the rising intermodel spread in the western part of the potential nondesert zone can be easily caught, especially under SSP5–8.5. This indicates that at the end of this century, the ensemble members will show less agreement in their projections of the nondesert type over this region.

d. The Sahara Desert’s responses to twenty-first-century climate change

The preceding analysis highlights the large intermodel spread in the Sahara Desert’s responses, in which different conclusions are observed among the ensemble members. Multimodel ensemble mean is often utilized in previous studies to handle such uncertainty. However, the large intermodel spread would be masked by such a method, leading to less robust ensemble results. A reliable ensemble method, i.e., EBDA, is therefore developed to reflect the Sahara’s future responses under uncertainty (see the data and methods section for detailed information on the presented method).

Variations in the area of the Sahara Desert, the nondesert zone, and the transition zone (approximately, the Sahel, investigated at the annual scale only) are reported in Table 3. Figure 6 displays the advance/retreat of the Sahara Desert in two future periods (compared with the period 1995–2014) under two climate scenarios. For SSP2–4.5, apart from some slight northwestern intrusions in Morocco and Algeria, the Sahara shows negligible variations in other regions over time. The area of the Sahara Desert reported in Table 3 also remains stable. This seems to be contrary to the results of the preceding uncertainty analysis; namely, more models support the Saharan northward shift by 2080–99. In fact, CMIP6 models show large intermodel spread (i.e., large standard deviation of probabilities generated from EBDA) over many grids, causing them to be statistically not distinguishable (color stippling in Fig. 6) regarding the three zone types (desert, transition, and nondesert zones). Thus, although there seem to be more models supporting the northward shift, the robust northward retreat of the Sahara Desert under SSP2–4.5 cannot be concluded from the statistical perspective. On the other hand, despite the significant uncertainties reported in section 3c, the area of the transition zone decreases significantly (310 622 km2, a 19% decrease from 1995–2014 to 2080–99; Fig. S14). Variations of the nondesert zone are not significant; the area shows subtle expansion (73 699 km2, a 1% increase). The total number of statistically indistinguishable grids, however, shows an evident increase over future periods (see Table 3). For SSP5–8.5, compared with the period 1995–2014, robust northward shifts of the Sahara Desert (i.e., northwestern expansion to Morocco and Algeria, and northward evacuation in Niger, Chad, and Sudan) are observable in both periods. For the western part of the southern boundary, however, it displays noticeable southward advances over Senegal and Mauritania under SSP5–8.5. Statistically indistinguishable grids mainly exist in the boundary as well as in East Africa. It is interesting to note that the number of such grids under SSP5–8.5 is much smaller than that in SSP2–4.5 or the historical period (Table 3).

Table 3

Simulated historical/future areas of three/two vegetation-based zones and the number of statistically indistinguishable grid points under SSP2–4.5 and SSP5–8.5.

Table 3
Fig. 6.
Fig. 6.

Projected responses of the Sahara Desert to climate change at the annual scale under (left) SSP2–4.5 and (right) SSP5–8.5. Brown (blue) boundary marks the Sahara Desert (nondesert zone). Khaki (green) shading indicates the advance (retreat) of the Sahara or the retreat (advance) of nondesert zone. Color stippling implies significant disagreement (p value > 0.05) such that the grids cannot be distinguished statistically; circular (triangular) stippling correspond to the historical (future) period(s). The darker the color, the higher the uncertainty level (i.e., higher p value).

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

Furthermore, in addition to the significant northward creeping of the nondesert zone, distinct eastern expansion over Kenya, Ethiopia, and Somalia is also revealed. Enormous increases in the area of the nondesert zone over the two future periods are reported in Table 3. By the end of this century, the expansion area (401 486 km2) would be about a quarter of the transition zone under SSP5–8.5. In addition, compared with the period 1995–2014, a significant shrinking in the area of the transition zone is found over 2080–99 under the same scenario (203 003 km2, a 12% decrease). Evident northward shifts of the Sahel are visible in Fig. S14. Hefty footprints of the Sahara over Morocco and Algeria also characterize the shrinking of the transition zone in the northwest of North Africa.

The responses of the Sahara Desert to climate change may be underestimated due to taking the annual mean of precipitation; advance/retreat of the Sahara may be more distinct at the seasonal scale, which is of great importance for long-term water resource management in North Africa. To provide insight into the seasonal responses to climate change, the future evolution of the Sahara Desert is also analyzed at wet- and dry-season scales (see Figs. 7 and 8, and Table 3). Under both SSPs, apart from the Sahara’s notable retreat in Somalia and advance in the northwest (more significant in SSP5–8.5), the variations of the Sahara Desert in the southern boundary are almost negligible in the dry season. Moreover, the remarkable shrinking over the west of the nondesert zone can be found under both scenarios. This retreat will mainly impact Liberia, Côte d’Ivoire, and Ghana. By contrast, northward and eastward expansion of the nondesert zone occur over eastern regions under both scenarios (especially in 2080–99), albeit with some not statistically significant grids. Overall, models under the two scenarios exhibit similar but varying degrees of changes in the Saharan future variations.

Fig. 7.
Fig. 7.

As in Fig. 6, but for dry-season responses of desert and nondesert zones.

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

Fig. 8.
Fig. 8.

As in Fig. 6, but for wet-season responses of desert and nondesert zones.

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

The wet-season responses tell a slightly different story. Under SSP2–4.5, the changes in the western part, such as in Senegal and Mali, are trivial. Nevertheless, the slight northward retreat in the middle and eastern end of the Sahara’s southern boundary is clear in Fig. 8. Considerable decreases in the Sahara gross area are also reported in Table 3, while the number of statistically indistinguishable grid points increases over future periods. This highlights the superiority of the developed EBDA method; it can not only provide insight into robust responses of the Sahara Desert but also reveal the variations in the intermodel spread. In addition, more evident responses of the Sahara Desert, namely, the southward advance to Senegal and Mali, as well as the northward migration over Niger, Chad, and Sudan, can be found in SSP5–8.5. Compared with the historical period, the Sahara’s retreat in the wet season under this scenario is more significant than that under SSP2–4.5, with its area decreasing by 724 236 km2 at the end of this century. Moreover, eastward intrusion of the nondesert zone can easily be noted, although it is not as significant as that at the dry-season scale. By contrast, the whole boundary’s northward advance is more noticeable in the wet season, which may benefit many countries’ socioeconomic development (e.g., Burkina Faso, Nigeria, Chad, and Sudan). Over East Africa, the number of statistically indistinguishable grids increases substantially compared with that in the historical period. This is accompanied by significant retreat of the Sahara Desert in this region (17% and 11% decreases of the area by 2080–99 under SSP2–4.5 and SSP5–8.5, respectively). In general, the future evolution of the Sahara Desert and the comparison with its historical state consistently suggest the Sahara and Sahel’s gradual northward shifts and the nondesert zone’s eastern expansion, which is robust across different climate models. This pattern appears to be decomposed into two single patterns in wet and dry seasons with an amplified degree of variations. The northward advance of the Saharan northwestern boundary and eastern expansion of the nondesert zone are mainly reflected in the dry season. The more noticeable northward retreat (advance) of the Sahara Desert (the nondesert zone) mainly appear in the wet season. Meanwhile, the unexpected expansion of the Sahara over part of West Africa is also more evident compared with changes at the annual scale. The developed method reveals not only robust annual/seasonal responses of the Sahara Desert to climate change but also variations in the intermodel uncertainty. The documented drying in the Mediterranean and wetting in the Sahel (Almazroui et al. 2020; Monerie et al. 2020; Li et al. 2021) can both justify one of our findings (i.e., northward shifts of the Sahara Desert), which further demonstrates the robustness of the developed EBDA.

The antecedent analysis reveals distinct responses of the Sahara Desert to climate change at seasonal/annual scales. It also shows that many grids remain statistically indistinguishable despite the developed EBDA; the projections among individual models are significantly different. Besides, the number of such grids is substantially different between two SSPs (Table 3), indicating that the uncertainties in the projections are related not only to how models reflect the climate system but also to different forcing scenarios. In addition, different GCM/SSP combinations may bring about different zone-classification results (i.e., different projected responses of the Sahara) due to their interactive effects. Thus, factorial multivariate analysis of variance and covariance (MANOVA) is developed to investigate whether the choices of CMIP6 models (GCMs) and forcing scenarios (SSPs), as well as their interaction (GCM × SSP), would exert significant effects on projections of the Saharan responses to climate change. The detailed descriptions for MANOVA can be found in section 2 of the supplemental material.

With the established factorial MANOVA (considering all probabilistic information of the preclassified three/two zones from EBDA as responses), the level of statistical significance (p value) for each source at each grid is estimated. When the p value is smaller than the predefined significance level (i.e., 0.05 in this study) at a certain grid, the corresponding factor (i.e., GCM, SSP, or GCM × SSP) would be considered as a significant one for leading to the spread in BDA results. For instance, if p < 0.05 for GCM × SSP at a given grid, the interactive effects of GCM and SSP on the BDA results (and thus the identification of the Sahara Desert) are significant.

As expected, the CMIP6 models have substantial impacts on the projected Saharan responses through BDA; the term is statistically significant at every grid point (not shown in figures). Figure 9 displays the remarkable spatial variations in statistical significance (p value) of other contributing factors to the uncertainties. For the annual scale, the forcing scenarios exhibit significant influences on the uncertainty in projections, particularly over low-latitude regions (0°–15°N). This is particularly true for regions with abundant rainfall (e.g., the Sahel as well as northwest and east of North Africa) since the differences in precipitation under two SSPs can significantly affect the projections of the Saharan responses. Even with the developed EBDA method, some grids (e.g., in East Africa) are still statistically indistinguishable owing to the high uncertainty caused by SSP. On the other hand, the statistical significance of the interaction term also shows evident spatial variability. It is not significant over the northwestern and eastern parts, which is completely different from the spatial pattern of SSP. The most significant grid cells (with the smallest p value) are concentrated in West Africa, with their uncertainties well handled by the EBDA (see Fig. 6).

Fig. 9.
Fig. 9.

Spatial distribution of the significance levels (p value < 0.05) for (left) GCM × SSP and (right) SSP.

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

For the dry-season scale, the number of significant grids decreases substantially (Fig. 9). The interaction term is significant over regions located around 10°N; the different SSPs put their specific imprint among GCMs. Note that some grids there are classified as statistically indistinguishable ones through EBDA (see Fig. 7), suggesting that, apart from the main effects of GCM and SSP, their interactions show significant indirect effects on the projected Saharan responses. Furthermore, the sub-Saharan region has limited rainfall, especially in the dry season; therefore, the difference in precipitation projections under two SSPs is trivial and thus may not substantially affect the projections. Consequently, the SSP term in most grids over this region is deemed an insignificant impact factor. By contrast, significant effects of the SSP term can be found in a large number of grids in the wet season when substantial rainfall appears over the vast area of North Africa. Moreover, the interaction term is identified as a significant factor in many grids within 20°–30°N, which is not evident at the dry-season scale.

Overall, although significant effects of GCM, SSP, and their interaction are identified in many grids, most of the uncertainties caused by these factors could be well handled by the developed EBDA method. The forcing scenarios and CMIP6 models are the main sources of uncertainty over regions where grids are often considered statistically indistinguishable. In addition, the indirect effects on uncertainty owing to their interactions are particularly significant at the wet-season scale.

e. Potential mechanisms for northward shifts of the Sahara Desert

Significant northward shifts of the Sahara Desert are revealed at the annual scale (i.e., the northward retreat of the southern boundary and the northward advance of the northwestern one). In this section, the underlying mechanism for such retreat and advance will be discussed for wet and dry seasons, respectively.

For the wet season, the projected retreat of the Sahara Desert may be induced by northward migration of the intertropical convergence zone (ITCZ), a result of intensified surface warming over the Sahara (Mamalakis et al. 2021). In detail, under both scenarios of SSP2–4.5 and SSP5–8.5, significant norward shifts of ITCZ in response to climate change can be identified, as shown in Figs. S15a and S15b, particularly for the wet season. This could be related to the fact that the strengthening Saharan heat low (SHL) (as shown in Figs. S15c,d) associated with significantly increased surface temperature and reduced geopotential height (as shown in Figs. 10a,c) could result in farther northward migration of ITCZ during the wet season. This is consistent with the previous finding (Dunning et al. 2018, using CMIP5 models). The high correlation between the changes in the SHL index and those of the ITCZ position (Fig. S15e) further supports our postulation. In addition, surface warming over the Sahara (lower surface pressure) favors moisture transport toward the Sahel region, which triggers a positive anomaly in RH, as shown in Fig. 10e. This would make ITCZ reside there longer (Dunning et al. 2018). On the other hand, a monsoon trough around 20°N as a typical circulation feature during the West African monsoon season (WAM) is displayed in Fig. S16a. A positive anomaly in 925-hPa geopotential height over the south of 20°N and a negative one over the north of 20°N (Fig. S16b) indicate a northward shift of the monsoon trough during WAM (Vizy and Cook 2017). Such a shift is accompanied by anomalous westerly and southwesterly moisture flux from the Gulf of Guinea to West Africa (Fig. 10e). These patterns may alter the seasonality of precipitation and induce increased rainfall, leading to the projected retreat of the southern boundary, albeit with significant intermodeling uncertainties.

Fig. 10.
Fig. 10.

Variations in ensemble means of (a) absolute and (b) fractional surface temperature change (°C and %, respectively) in wet and dry seasons, (c) geopotential heights (m) over June–August (JJA), (d) total cloud-cover percentage in the dry season, and (e), (f) 925-hPa relative humidity (%) and moisture flux (g kg−1 m s−1) in wet and dry seasons of 2080–99 (SSP5–8.5) vs those of 1995–2014. Note that for (a)–(d), black dots denote significant changes at 95% confidence level, and for (c), JJA mean of geopotential heights is selected for demonstrating the intensified SHL.

Citation: Journal of Climate 36, 10; 10.1175/JCLI-D-22-0169.1

It is well known that land surface shows more rapid warming than surrounding oceans (Lambert and Chiang 2007; Sutton et al. 2007). Figure 10b shows the fractional changes in surface temperature over North Africa in the dry season, suggesting an increasing trend of land–ocean contrast. This may lead to an amplified drying trend in the northwest where precipitation is largely affected by moisture transports from the surrounding ocean (Barcikowska et al. 2018). Specifically, the increase in saturation-specific humidity qs with temperature may lead to a decreased lapse rate over the sea ΓO (Holton 1973), while the lapse rate over land ΓL is closer to the dry adiabatic lapse rate (a constant that is independent of qs). Thus, ΓL will decrease less than ΓO with increasing qs (Joshi et al. 2008). Such a difference will augment the trend of lower-tropospheric temperature over land, leading to a substantial increase in qs based on the Clausius–Clapeyron relationship (Joshi et al. 2008). However, less warming over the ocean may in turn constrain the rise in qs above it and further limit moisture supply to the land (Fasullo 2010). The increased qs level over the land and reduced incoming moisture (from the ocean) would trigger evident decreases in RH over the northwest of North Africa (Fig. 10f), which subsequently reduces the cloud cover percentage (Fig. 10d). The decreased RH and cloud cover over the northwest of the Sahara may then result in decreased rainfall there. Moreover, an expansion of the Hadley cell is projected by the selected 10 CMIP6 models (Fig. S17), which is consistent with the results of Johanson and Fu (2009). Previous research has suggested a close relationship between the widening of the tropics and the expansion of the Sahara (Thomas and Nigam 2018). Similarly, the northward shift of the poleward edge for the Hadley cell is also found to demonstrate high correlations (Fig. S16c) with the projected advance of the Sahara’s northwest, especially under SSP5–8.5 (r = 0.74). Thus, the land–ocean contrast and the expansion of the Hadley cell may intensify the aridity in the northwest of North Africa, resulting in the northward advance of the Sahara’s northwestern boundary.

4. Conclusions and discussion

Uncertainties over the regions are shown through the respective frequency (which is based on a subset of 10 best-performing CMIP6 models) of the determined three (desert, transition, and nondesert) or two (desert and nondesert) vegetation-based zones. An ensemble Bayesian discriminant analysis approach (EDBA), differing from the previously simple one (i.e., using precipitation isoline), is developed to handle such uncertainties and further create robust projections of the Sahara’s ensemble responses to climate change. Moreover, factorial MANOVA is undertaken to investigate whether different climate models, forcing scenarios, and their interaction would demonstrate significant effects on the projected responses of the Sahara Desert. The resulting findings include the following:

  • Northward shifts of both the Sahara and Sahel, as well as eastern expansion of the nondesert zone, are identified at the annual scale, which is statistically significant in the 10 best-performing ensemble members.

  • The pattern of variations at the annual scale is reflected in wet (the northward retreat of the Sahara Desert and the northward advance of the nondesert zone) and dry (the northward advance of the Saharan northwestern boundary and the eastern expansion of the nondesert zone) seasons.

  • The factors of climate model and forcing scenarios have not only direct effects from the factors themselves on zone classification but also indirect ones due to their interaction. Most of these complexities are robustly handled through the proposed EBDA, which further highlights the superiority of the developed approach.

Furthermore, variations in meteorological systems under climate change are investigated to explore plausible mechanisms for the northward shifts of the Sahara Desert. In detail, for the wet season, the intensified SHL (lower surface pressure and geopotential height) will make ITCZ step farther north and stay there longer. A northward shift of the monsoon trough, accompanied by anomalous westerly/southwesterly moisture flux from the Gulf of Guinea to West Africa, is also found during WAM. These patterns will trigger increased precipitation over the Sahel and further lead to the projected retreat of the Sahara’s southern boundary. For the dry season, the difference in the lapse rate between land and ocean would contribute to increased contrast in land versus ocean warming over the future period. Such contrast will induce reduced RH and cloud cover over the northwest of North Africa and thus decreased precipitation there, which may cause the expansion of the Sahara’s northwest. Significant correlations between the poleward expansion of the Hadley cell and the projected northward advance of the Sahara’s northwestern boundary are also revealed. Further analysis is desired owing to uncertainties in the considered driving factors among multiple climate models. Future work is also needed to provide a more complete understanding of potential mechanisms through a systematic experimental design and comprehensive examination of the atmospheric moisture budget. Additionally, previous studies reported close relationships between the variations in sea surface temperatures over different oceanic basins and the interannual/decadal variability in precipitation over North Africa (Rodríguez-Fonseca et al. 2015; Pausata et al. 2020; Stige et al. 2006). For instance, several studies have indicated that increased positive Indian Ocean dipole (Saji et al. 1999; Evan et al. 2015) might cause amplified wind and thermocline changes and thus trigger enhanced short rains (in boreal autumn) over East Africa (Black et al. 2003; Shongwe et al. 2011; Dunning et al. 2018); besides, humid air induced by the enhanced evaporation could advect from the Mediterranean to the Sahara, resulting in amplified low-level moisture convergence and thus increased rainfall over the Sahel (Park et al. 2016). A follow-on analysis for potential links between the SST variations and the projected responses of the Sahara Desert would be an interesting extension of this research.

The results of our study would help advance the understanding of the Sahara Desert’s responses to the projected climate change. Compared with simple precipitation isolines used in previous studies, the developed EBDA can build multidimensional interrelationships between vegetation cover and precipitation level. Based on such relationships, the study region can be classified into multiple vegetation-based zones. The differences among multiple climate models within the CMIP6 ensemble are then analyzed through the comparisons of posterior probabilities obtained from BDA, to address the effects of modeling uncertainties and generate robust projections. Although the developed approach can address most of the uncertainties induced by climate models, forcing scenarios, and their interaction, there are still a few grids that cannot be distinguished statistically. This might be caused by the insufficient characterization of the vegetation’s spatial variability that relies on a single discriminant variable (i.e., precipitation). A multivariate BDA is thus desired in our future work to further improve the robustness in characterizing the spatial heterogeneity of vegetation.

Acknowledgments.

This research was supported by the Canada Research Chair Program, the Natural Science and Engineering Research Council of Canada, Western Economic Diversification (15269), and MITACS. We are also very grateful for the helpful inputs from the editor and anonymous reviewers.

Data availability statement.

Data analyzed in this research were obtained from the following publicly available datasets: the Coupled Model Intercomparison Project (Phase 6) of the World Climate Research Programme, available at https://esgf-node.llnl.gov/projects/cmip6/; NOAA Climate Data Record of Normalized Difference Vegetation Index, available at https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.ncdc:C00813; Global Precipitation Climatology Centre dataset, available at https://opendata.dwd.de/climate_environment/GPCC/html/gpcc_monitoring_v6_doi_download.html.

REFERENCES

  • Abdi, H., and L. J. Williams, 2010: Tukey’s honestly significant difference (HSD) test. Encyclopedia of Research Design, N. J. Salkind, Ed., Sage Publications, 1566–1570, https://doi.org/10.4135/9781412961288.

  • Almazroui, M., F. Saeed, S. Saeed, M. Nazrul Islam, M. Ismail, N. A. B. Klutse, and M. H. Siddiqui, 2020: Projected change in temperature and precipitation over Africa from CMIP6. Earth Syst. Environ., 4, 455475, https://doi.org/10.1007/s41748-020-00161-x.

    • Search Google Scholar
    • Export Citation
  • Ansell, L., and L. D. Valle, 2021: Social media integration of flood data: A vine copula-based approach. arXiv, 2104.01869v2, https://doi.org/10.48550/arXiv.2104.01869.

  • Anyamba, A., J. L. Small, C. J. Tucker, and E. W. Pak, 2014: Thirty-two years of Sahelian zone growing season non-stationary NDVI3g patterns and trends. Remote Sens., 6, 31013122, https://doi.org/10.3390/rs6043101.

    • Search Google Scholar
    • Export Citation
  • Babaousmail, H., R. Hou, B. Ayugi, M. Ojara, H. Ngoma, R. Karim, A. Rajasekar, and V. Ongoma, 2021: Evaluation of the performance of CMIP6 models in reproducing rainfall patterns over North Africa. Atmosphere, 12, 475, https://doi.org/10.3390/atmos12040475.

    • Search Google Scholar
    • Export Citation
  • Bação, F., V. Lobo, and M. Painho, 2005: Self-organizing maps as substitutes for k-means clustering. Computational Science—ICCS 2005, V. S. Sunderam et al., Eds., Lecture Notes in Computer Science, Vol. 3516, Springer, 476–483.

  • Bagan, H., Q. Wang, M. Watanabe, Y. Yang, and J. Ma, 2005: Land cover classification from MODIS EVI times-series data using SOM neural network. Int. J. Remote Sens., 26, 49995012, https://doi.org/10.1080/01431160500206650.

    • Search Google Scholar
    • Export Citation
  • Barcikowska, M. J., S. B. Kapnick, and F. Feser, 2018: Impact of large-scale circulation changes in the North Atlantic sector on the current and future Mediterranean winter hydroclimate. Climate Dyn., 50, 20392059, https://doi.org/10.1007/s00382-017-3735-5.

    • Search Google Scholar
    • Export Citation
  • Battisti, D. S., and R. L. Naylor, 2009: Historical warnings of future food insecurity with unprecedented seasonal heat. Science, 323, 240244, https://doi.org/10.1126/science.1164363.

    • Search Google Scholar
    • Export Citation
  • Becker, A., P. Finger, A. Meyer-Christoffer, B. Rudolf, K. Schamm, U. Schneider, and M. Ziese, 2013: A description of the global land-surface precipitation data products of the Global Precipitation Climatology Centre with sample applications including centennial (trend) analysis from 1901–present. Earth Syst. Sci. Data, 5, 7199, https://doi.org/10.5194/essd-5-71-2013.

    • Search Google Scholar
    • Export Citation
  • Bigdeli, A., A. Maghsoudi, and R. Ghezelbash, 2022: Application of self-organizing map (SOM) and k-means clustering algorithms for portraying geochemical anomaly patterns in Moalleman district, NE Iran. J. Geochem. Explor., 233, 106923, https://doi.org/10.1016/j.gexplo.2021.106923.

    • Search Google Scholar
    • Export Citation
  • Black, E., J. Slingo, and K. R. Sperber, 2003: An observational study of the relationship between excessively strong short rains in coastal East Africa and Indian Ocean SST. Mon. Wea. Rev., 131, 7494, https://doi.org/10.1175/1520-0493(2003)131<0074:AOSOTR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Boko, M., and Coauthors, 2007: Africa. Climate Change 2007: Impacts, Adaptation and Vulnerability, M. L. Parry et al., Eds., Cambridge University Press, 433–467, https://www.ipcc.ch/site/assets/uploads/2018/02/ar4-wg2-chapter9-1.pdf.

  • Burke, M. B., D. B. Lobell, and L. Guarino, 2009: Shifts in African crop climates by 2050, and the implications for crop improvement and genetic resources conservation. Global Environ. Change, 19, 317325, https://doi.org/10.1016/j.gloenvcha.2009.04.003.

    • Search Google Scholar
    • Export Citation
  • Chen, W., Z. Jiang, and L. Li, 2011: Probabilistic projections of climate change over China under the SRES A1B scenario using 28 AOGCMs. J. Climate, 24, 47414756, https://doi.org/10.1175/2011JCLI4102.1.

    • Search Google Scholar
    • Export Citation
  • Chou, C., J. C. H. Chiang, C.-W. Lan, C.-H. Chung, Y.-C. Liao, and C.-J. Lee, 2013: Increase in the range between wet and dry season precipitation. Nat. Geosci., 6, 263267, https://doi.org/10.1038/ngeo1744.

    • Search Google Scholar
    • Export Citation
  • Cook, K. H., and E. K. Vizy, 2015: Detection and analysis of an amplified warming of the Sahara Desert. J. Climate, 28, 65606580, https://doi.org/10.1175/JCLI-D-14-00230.1.

    • Search Google Scholar
    • Export Citation
  • Dong, B., and R. Sutton, 2015: Dominant role of greenhouse-gas forcing in the recovery of Sahel rainfall. Nat. Climate Change, 5, 757760, https://doi.org/10.1038/nclimate2664.

    • Search Google Scholar
    • Export Citation
  • Droogers, P., W. W. Immerzeel, W. Terink, J. Hoogeveen, M. F. P. Bierkens, L. P. H. Van Beek, and B. Debele, 2012: Water resources trends in Middle East and North Africa towards 2050. Hydrol. Earth Syst. Sci., 16, 31013114, https://doi.org/10.5194/hess-16-3101-2012.

    • Search Google Scholar
    • Export Citation
  • Dunning, C. M., E. Black, and R. P. Allan, 2018: Later wet seasons with more intense rainfall over Africa under future climate change. J. Climate, 31, 97199738, https://doi.org/10.1175/JCLI-D-18-0102.1.

    • Search Google Scholar
    • Export Citation
  • Evan, A. T., C. Flamant, C. Lavaysse, C. Kocha, and A. Saci, 2015: Water vapor–forced greenhouse warming over the Sahara Desert and the recent recovery from the Sahelian drought. J. Climate, 28, 108123, https://doi.org/10.1175/JCLI-D-14-00039.1.

    • Search Google Scholar
    • Export Citation
  • Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 19371958, https://doi.org/10.5194/gmd-9-1937-2016.

    • Search Google Scholar
    • Export Citation
  • Fasullo, J. T., 2010: Robust land–ocean contrasts in energy and water cycle feedbacks. J. Climate, 23, 46774693, https://doi.org/10.1175/2010JCLI3451.1.

    • Search Google Scholar
    • Export Citation
  • Folland, C. K., T. N. Palmer, and D. E. Parker, 1986: Sahel rainfall and worldwide sea temperatures, 1901–85. Nature, 320, 602607, https://doi.org/10.1038/320602a0.

    • Search Google Scholar
    • Export Citation
  • Girden, E. R., 1992: ANOVA: Repeated Measures. Sage Publications, 79 pp.

  • Hastie, T., R. Tibshirani, and J. Friedman, 2009: The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer, 745 pp.

  • Herrmann, S. M., A. Anyamba, and C. J. Tucker, 2005: Recent trends in vegetation dynamics in the African Sahel and their relationship to climate. Global Environ. Change, 15, 394404, https://doi.org/10.1016/j.gloenvcha.2005.08.004.

    • Search Google Scholar
    • Export Citation
  • Holton, J. R., 1973: An introduction to dynamic meteorology. Amer. J. Phys., 41, 752754, https://doi.org/10.1119/1.1987371.

  • Hu, J., L. Sun, C. H. Li, X. Wang, X. L. Jia, and Y. P. Cai, 2018: Water quality risk assessment for the Laoguanhe River of China using a stochastic simulation method. J. Environ. Inf., 31, 123136, https://doi.org/10.3808/jei.201800387.

    • Search Google Scholar
    • Export Citation
  • Huang, W. W., X. J. Chen, Y. R. Fan, and Y. P. Li, 2022: Management of drinking water source in rural communities under climate change. J. Environ. Inf., 39, 136151, https://doi.org/10.3808/jei.202000431.

    • Search Google Scholar
    • Export Citation
  • Johanson, C. M., and Q. Fu, 2009: Hadley cell widening: Model simulations versus observations. J. Climate, 22, 27132725, https://doi.org/10.1175/2008JCLI2620.1.

    • Search Google Scholar
    • Export Citation
  • Joshi, M. M., J. M. Gregory, M. J. Webb, D. M. H. Sexton, and T. C. Johns, 2008: Mechanisms for the land/sea warming contrast exhibited by simulations of climate change. Climate Dyn., 30, 455465, https://doi.org/10.1007/s00382-007-0306-1.

    • Search Google Scholar
    • Export Citation
  • Kim, D., M. Chin, L. A. Remer, T. Diehl, H. Bian, H. Yu, M. E. Brown, and W. R. Stockwell, 2017: Role of surface wind and vegetation cover in multi-decadal variations of dust emission in the Sahara and Sahel. Atmos. Environ., 148, 282296, https://doi.org/10.1016/j.atmosenv.2016.10.051.

    • Search Google Scholar
    • Export Citation
  • Kohonen, T., 1982: Self-organized formation of topologically correct feature maps. Biol. Cybern., 43, 5969, https://doi.org/10.1007/BF00337288.

    • Search Google Scholar
    • Export Citation
  • Kucuk, U., M. Eyuboglu, H. O. Kucuk, and G. Degirmencioglu, 2016: Importance of using proper post hoc test with ANOVA. Int. J. Cardiol., 209, 346, https://doi.org/10.1016/j.ijcard.2015.11.061.

    • Search Google Scholar
    • Export Citation
  • Lambert, F. H., and J. C. H. Chiang, 2007: Control of land-ocean temperature contrast by ocean heat uptake. Geophys. Res. Lett., 34, L13704, https://doi.org/10.1029/2007GL029755.

    • Search Google Scholar
    • Export Citation
  • Lamjiak, T., J. Polvichai, and P. Varnakovida, 2016: A geometrical data classification using self-organizing map with fixed possible matching units. 2016 Int. Computer Science and Engineering Conf. (ICSEC), Chiang Mai, Thailand, Institute of Electrical and Electronics Engineers, 1–6, https://doi.org/10.1109/ICSEC.2016.7859914.

  • Lasaponara, R., A. M. Proto, A. Aromando, G. Cardettini, V. Varela, and M. Danese, 2020: On the mapping of burned areas and burn severity using self organizing map and sentinel-2 data. IEEE Geosci. Remote Sens. Lett., 17, 854858, https://doi.org/10.1109/LGRS.2019.2934503.

    • Search Google Scholar
    • Export Citation
  • Li, H., H. Wang, and Y. Yin, 2012: Interdecadal variation of the West African summer monsoon during 1979–2010 and associated variability. Climate Dyn., 39, 28832894, https://doi.org/10.1007/s00382-012-1426-9.

    • Search Google Scholar
    • Export Citation
  • Li, H., Z. Li, Y. Chen, Y. Xiang, Y. Liu, P. M. Kayumba, and X. Li, 2021: Drylands face potential threat of robust drought in the CMIP6 SSPs scenarios. Environ. Res. Lett., 16, 114004, https://doi.org/10.1088/1748-9326/ac2bce.

    • Search Google Scholar
    • Export Citation
  • Li, T. H., and W. C. Xie, 2021: A new method for computing the sediment delivery ratio for the hyper-concentrated flow areas of the Loess Plateau, China. J. Environ. Inf., 39, 110, https://doi.org/10.3808/jei.202100456.

    • Search Google Scholar
    • Export Citation
  • Liu, J., S. Fritz, C. F. A. Van Wesenbeeck, M. Fuchs, L. You, M. Obersteiner, and H. Yang, 2008: A spatially explicit assessment of current and future hotspots of hunger in sub-Saharan Africa in the context of global change. Global Planet. Change, 64, 222235, https://doi.org/10.1016/j.gloplacha.2008.09.007.

    • Search Google Scholar
    • Export Citation
  • Liu, Y., and Y. Xue, 2020: Expansion of the Sahara Desert and shrinking of frozen land of the Arctic. Sci. Rep., 10, 4109, https://doi.org/10.1038/s41598-020-61085-0.

    • Search Google Scholar
    • Export Citation
  • Liu, Y. L., J. Z. Du, Q. Wang, W. Yang, and B. S. Cui, 2022: Toward an assessment of runoff and thermal connectivity in a river-lake system within an urban environment. J. Environ. Inf., 40, 125140, https://doi.org/10.3808/jei.202200482.

    • Search Google Scholar
    • Export Citation
  • Lu, W., X. S. Qin, and C. Jun, 2017: A parsimonious framework of evaluating WSUD features in urban flood mitigation. J. Environ. Inf., 33, 1727, https://doi.org/10.3808/jei.201700373.

    • Search Google Scholar
    • Export Citation
  • Maidment, R. I., R. P. Allan, and E. Black, 2015: Recent observed and simulated changes in precipitation over Africa. Geophys. Res. Lett., 42, 81558164, https://doi.org/10.1002/2015GL065765.

    • Search Google Scholar
    • Export Citation
  • Mamalakis, A., and Coauthors, 2021: Zonally contrasting shifts of the tropical rain belt in response to climate change. Nat. Climate Change, 11, 143151, https://doi.org/10.1038/s41558-020-00963-x.

    • Search Google Scholar
    • Export Citation
  • Monerie, P.-A., C. M. Wainwright, M. Sidibe, and A. A. Akinsanola, 2020: Model uncertainties in climate change impacts on Sahel precipitation in ensembles of CMIP5 and CMIP6 simulations. Climate Dyn., 55, 13851401, https://doi.org/10.1007/s00382-020-05332-0.

    • Search Google Scholar
    • Export Citation
  • Niang, I., and Coauthors, 2014: Africa. Climate Change 2014: Impacts, Adaptation, and Vulnerability. Part B: Regional Aspects, V. R. Barros et al., Eds., Cambridge University Press, 1199–1265, https://www.ipcc.ch/site/assets/uploads/2018/02/WGIIAR5-Chap22_FINAL.pdf.

  • Nicholson, S. E., A. K. Dezfuli, and D. Klotter, 2012: A two-century precipitation dataset for the continent of Africa. Bull. Amer. Meteor. Soc., 93, 12191231, https://doi.org/10.1175/BAMS-D-11-00212.1.

    • Search Google Scholar
    • Export Citation
  • Notter, B., H. Hurni, U. Wiesmann, and J. O. Ngana, 2013: Evaluating watershed service availability under future management and climate change scenarios in the Pangani Basin. Phys. Chem. Earth, 6162, 111, https://doi.org/10.1016/j.pce.2012.08.017.

    • Search Google Scholar
    • Export Citation
  • Pan, Y., X. K. Zeng, X. Y. Gao, H. X. Xu, Y. Y. Sun, D. Wang, and J. C. Wu, 2022: Assessing human health risk to DNAPLs exposure in Bayesian uncertainty analysis. J. Environ. Inf., 39, 6780, https://doi.org/10.3808/jei.202100460.

    • Search Google Scholar
    • Export Citation
  • Park, J.-Y., J. Bader, and D. Matei, 2016: Anthropogenic Mediterranean warming essential driver for present and future Sahel rainfall. Nat. Climate Change, 6, 941945, https://doi.org/10.1038/nclimate3065.

    • Search Google Scholar
    • Export Citation
  • Pausata, F. S. R., M. Gaetani, G. Messori, A. Berg, D. M. de Souza, R. F. Sage, and P. B. DeMenocal, 2020: The greening of the Sahara: Past changes and future implications. One Earth, 2, 235250, https://doi.org/10.1016/j.oneear.2020.03.002.

    • Search Google Scholar
    • Export Citation
  • Pérez-Hoyos, A., B. Martínez, F. J. García-Haro, Á. Moreno, and M. A. Gilabert, 2014: Identification of ecosystem functional types from coarse resolution imagery using a self-organizing map approach: A case study for Spain. Remote Sens., 6, 1139111419, https://doi.org/10.3390/rs61111391.

    • Search Google Scholar
    • Export Citation
  • Razali, N. M., and Y. B. Wah, 2011: Power comparisons of Shapiro-Wilk, Kolmogorov-Smirnov, Lilliefors and Anderson-Darling tests. J. Stat. Model. Anal., 2, 2133.

    • Search Google Scholar
    • Export Citation
  • Rodger, R. S., 1975: The number of non-zero, post hoc contrasts from ANOVA and error‐rate. I. Br. J. Math. Stat. Psychol., 28, 7178, https://doi.org/10.1111/j.2044-8317.1975.tb00549.x.

    • Search Google Scholar
    • Export Citation
  • Rodríguez-Fonseca, B., and Coauthors, 2015: Variability and predictability of West African droughts: A review on the role of sea surface temperature anomalies. J. Climate, 28, 40344060, https://doi.org/10.1175/JCLI-D-14-00130.1.

    • Search Google Scholar
    • Export Citation
  • Rowell, D. P., 2001: Teleconnections between the tropical Pacific and the Sahel. Quart. J. Roy. Meteor. Soc., 127, 16831706, https://doi.org/10.1002/qj.49712757512.

    • Search Google Scholar
    • Export Citation
  • Saji, N. H., B. N. Goswami, P. N. Vinayachandran, and T. Yamagata, 1999: A dipole mode in the tropical Indian Ocean. Nature, 401, 360363, https://doi.org/10.1038/43854.

    • Search Google Scholar
    • Export Citation
  • Schneider, U., A. Becker, P. Finger, A. Meyer-Christoffer, B. Rudolf, and M. Ziese, 2011: GPCC monitoring product: Near real-time monthly land-surface precipitation from rain-gauges based on SYNOP and CLIMAT data. DWD, accessed 3 January 2022, http://doi.org/10.5676/DWD_GPCC/MP_M_V5_100.

  • Schneider, U., A. Becker, P. Finger, A. Meyer-Christoffer, M. Ziese, and B. Rudolf, 2014: GPCC’s new land surface precipitation climatology based on quality-controlled in situ data and its role in quantifying the global water cycle. Theor. Appl. Climatol., 115, 1540, https://doi.org/10.1007/s00704-013-0860-x.

    • Search Google Scholar
    • Export Citation
  • Schuster-Wallace, C. J., S. E. Dickson-Anderson, S. M. Papalexiou, and A. El Ganzouri, 2022: Design and application of the Tank Simulation Model (TSM): Assessing the ability of rainwater harvesting to meet domestic water demand. J. Environ. Inf., 40, 1629, https://doi.org/10.3808/jei.202200477.

    • Search Google Scholar
    • Export Citation
  • Sharma, S., J. Waldman, S. Afshari, and B. Fekete, 2019: Status, trends and significance of American hydropower in the changing energy landscape. Renewable Sustainable Energy Rev., 101, 112122, https://doi.org/10.1016/j.rser.2018.10.028.

    • Search Google Scholar
    • Export Citation
  • Shongwe, M. E., G. J. van Oldenborgh, B. van den Hurk, and M. van Aalst, 2011: Projected changes in mean and extreme precipitation in Africa under global warming. Part II: East Africa. J. Climate, 24, 37183733, https://doi.org/10.1175/2010JCLI2883.1.

    • Search Google Scholar
    • Export Citation
  • Skonieczny, C., and Coauthors, 2019: Monsoon-driven Saharan dust variability over the past 240,000 years. Sci. Adv., 5, eaav1887, https://doi.org/10.1126/sciadv.aav1887.

    • Search Google Scholar
    • Export Citation
  • Srivastava, S., M. R. Gupta, and B. A. Frigyik, 2007: Bayesian quadratic discriminant analysis. J. Mach. Learn. Res., 8, 12771305.

  • Stige, L. C., J. Stave, K.-S. Chan, L. Ciannelli, N. Pettorelli, M. Glantz, H. R. Herren, and N. C. Stenseth, 2006: The effect of climate variation on agro-pastoral production in Africa. Proc. Natl. Acad. Sci. USA, 103, 30493053, https://doi.org/10.1073/pnas.0600057103.

    • Search Google Scholar
    • Export Citation
  • Student, 1908: The probable error of a mean. Biometrika, 6, 125, https://doi.org/10.2307/2331554.

  • Sun, Q., C. Miao, A. AghaKouchak, and Q. Duan, 2016: Century-scale causal relationships between global dry/wet conditions and the state of the Pacific and Atlantic Oceans. Geophys. Res. Lett., 43, 65286537, https://doi.org/10.1002/2016GL069628.

    • Search Google Scholar
    • Export Citation
  • Sutton, R. T., B. Dong, and J. M. Gregory, 2007: Land/sea warming ratio in response to climate change: IPCC AR4 model results and comparison with observations. Geophys. Res. Lett., 34, L02701, https://doi.org/10.1029/2006GL028164.

    • Search Google Scholar
    • Export Citation
  • Thomas, N., and S. Nigam, 2018: Twentieth-century climate change over Africa: Seasonal hydroclimate trends and Sahara Desert expansion. J. Climate, 31, 33493370, https://doi.org/10.1175/JCLI-D-17-0187.1.

    • Search Google Scholar
    • Export Citation
  • Titterington, D. M., A. F. M. Smith, and U. E. Makov, 1985: Statistical Analysis of Finite Mixture Distributions. Wiley, 243 pp.

  • Tucker, C. J., and S. E. Nicholson, 1999: Variations in the size of the Sahara Desert from 1980 to 1997. Ambio, 28, 587591.

  • Tucker, C. J., H. E. Dregne, and W. W. Newcomb, 1991: Expansion and contraction of the Sahara Desert from 1980 to 1990. Science, 253, 299300, https://doi.org/10.1126/science.253.5017.299.

    • Search Google Scholar
    • Export Citation
  • van der Does, M., and Coauthors, 2020: Tropical rains controlling deposition of Saharan dust across the North Atlantic Ocean. Geophys. Res. Lett., 47, e2019GL086867, https://doi.org/10.1029/2019GL086867.

    • Search Google Scholar
    • Export Citation
  • Viatte, G., J. De Graaf, M. Demeke, T. Takahatake, and M. R. de Arce, 2009: Responding to the food crisis: Synthesis of medium-term measures proposed in inter-agency assessments. Food Agriculture Organization Synthesis Rep., 102 pp., https://www.fao.org/fileadmin/user_upload/ISFP/SR_Web.pdf.

  • Vizy, E. K., and K. H. Cook, 2017: Seasonality of the observed amplified Sahara warming trend and implications for Sahel rainfall. J. Climate, 30, 30733094, https://doi.org/10.1175/JCLI-D-16-0687.1.

    • Search Google Scholar
    • Export Citation

Supplementary Materials

Save
  • Abdi, H., and L. J. Williams, 2010: Tukey’s honestly significant difference (HSD) test. Encyclopedia of Research Design, N. J. Salkind, Ed., Sage Publications, 1566–1570, https://doi.org/10.4135/9781412961288.

  • Almazroui, M., F. Saeed, S. Saeed, M. Nazrul Islam, M. Ismail, N. A. B. Klutse, and M. H. Siddiqui, 2020: Projected change in temperature and precipitation over Africa from CMIP6. Earth Syst. Environ., 4, 455475, https://doi.org/10.1007/s41748-020-00161-x.

    • Search Google Scholar
    • Export Citation
  • Ansell, L., and L. D. Valle, 2021: Social media integration of flood data: A vine copula-based approach. arXiv, 2104.01869v2, https://doi.org/10.48550/arXiv.2104.01869.

  • Anyamba, A., J. L. Small, C. J. Tucker, and E. W. Pak, 2014: Thirty-two years of Sahelian zone growing season non-stationary NDVI3g patterns and trends. Remote Sens., 6, 31013122, https://doi.org/10.3390/rs6043101.

    • Search Google Scholar
    • Export Citation
  • Babaousmail, H., R. Hou, B. Ayugi, M. Ojara, H. Ngoma, R. Karim, A. Rajasekar, and V. Ongoma, 2021: Evaluation of the performance of CMIP6 models in reproducing rainfall patterns over North Africa. Atmosphere, 12, 475, https://doi.org/10.3390/atmos12040475.

    • Search Google Scholar
    • Export Citation
  • Bação, F., V. Lobo, and M. Painho, 2005: Self-organizing maps as substitutes for k-means clustering. Computational Science—ICCS 2005, V. S. Sunderam et al., Eds., Lecture Notes in Computer Science, Vol. 3516, Springer, 476–483.

  • Bagan, H., Q. Wang, M. Watanabe, Y. Yang, and J. Ma, 2005: Land cover classification from MODIS EVI times-series data using SOM neural network. Int. J. Remote Sens., 26, 49995012, https://doi.org/10.1080/01431160500206650.

    • Search Google Scholar
    • Export Citation
  • Barcikowska, M. J., S. B. Kapnick, and F. Feser, 2018: Impact of large-scale circulation changes in the North Atlantic sector on the current and future Mediterranean winter hydroclimate. Climate Dyn., 50, 20392059, https://doi.org/10.1007/s00382-017-3735-5.

    • Search Google Scholar
    • Export Citation
  • Battisti, D. S., and R. L. Naylor, 2009: Historical warnings of future food insecurity with unprecedented seasonal heat. Science, 323, 240244, https://doi.org/10.1126/science.1164363.

    • Search Google Scholar
    • Export Citation
  • Becker, A., P. Finger, A. Meyer-Christoffer, B. Rudolf, K. Schamm, U. Schneider, and M. Ziese, 2013: A description of the global land-surface precipitation data products of the Global Precipitation Climatology Centre with sample applications including centennial (trend) analysis from 1901–present. Earth Syst. Sci. Data, 5, 7199, https://doi.org/10.5194/essd-5-71-2013.

    • Search Google Scholar
    • Export Citation
  • Bigdeli, A., A. Maghsoudi, and R. Ghezelbash, 2022: Application of self-organizing map (SOM) and k-means clustering algorithms for portraying geochemical anomaly patterns in Moalleman district, NE Iran. J. Geochem. Explor., 233, 106923, https://doi.org/10.1016/j.gexplo.2021.106923.

    • Search Google Scholar
    • Export Citation
  • Black, E., J. Slingo, and K. R. Sperber, 2003: An observational study of the relationship between excessively strong short rains in coastal East Africa and Indian Ocean SST. Mon. Wea. Rev., 131, 7494, https://doi.org/10.1175/1520-0493(2003)131<0074:AOSOTR>2.0.CO;2.

    • Search Google Scholar
    • Export Citation
  • Boko, M., and Coauthors, 2007: Africa. Climate Change 2007: Impacts, Adaptation and Vulnerability, M. L. Parry et al., Eds., Cambridge University Press, 433–467, https://www.ipcc.ch/site/assets/uploads/2018/02/ar4-wg2-chapter9-1.pdf.

  • Burke, M. B., D. B. Lobell, and L. Guarino, 2009: Shifts in African crop climates by 2050, and the implications for crop improvement and genetic resources conservation. Global Environ. Change, 19, 317325, https://doi.org/10.1016/j.gloenvcha.2009.04.003.

    • Search Google Scholar
    • Export Citation
  • Chen, W., Z. Jiang, and L. Li, 2011: Probabilistic projections of climate change over China under the SRES A1B scenario using 28 AOGCMs. J. Climate, 24, 47414756, https://doi.org/10.1175/2011JCLI4102.1.

    • Search Google Scholar
    • Export Citation
  • Chou, C., J. C. H. Chiang, C.-W. Lan, C.-H. Chung, Y.-C. Liao, and C.-J. Lee, 2013: Increase in the range between wet and dry season precipitation. Nat. Geosci., 6, 263267, https://doi.org/10.1038/ngeo1744.

    • Search Google Scholar
    • Export Citation
  • Cook, K. H., and E. K. Vizy, 2015: Detection and analysis of an amplified warming of the Sahara Desert. J. Climate, 28, 65606580, https://doi.org/10.1175/JCLI-D-14-00230.1.

    • Search Google Scholar
    • Export Citation
  • Dong, B., and R. Sutton, 2015: Dominant role of greenhouse-gas forcing in the recovery of Sahel rainfall. Nat. Climate Change, 5, 757760, https://doi.org/10.1038/nclimate2664.

    • Search Google Scholar
    • Export Citation
  • Droogers, P., W. W. Immerzeel, W. Terink, J. Hoogeveen, M. F. P. Bierkens, L. P. H. Van Beek, and B. Debele, 2012: Water resources trends in Middle East and North Africa towards 2050. Hydrol. Earth Syst. Sci., 16, 31013114, https://doi.org/10.5194/hess-16-3101-2012.

    • Search Google Scholar
    • Export Citation
  • Dunning, C. M., E. Black, and R. P. Allan, 2018: Later wet seasons with more intense rainfall over Africa under future climate change. J. Climate, 31, 97199738, https://doi.org/10.1175/JCLI-D-18-0102.1.

    • Search Google Scholar
    • Export Citation
  • Evan, A. T., C. Flamant, C. Lavaysse, C. Kocha, and A. Saci, 2015: Water vapor–forced greenhouse warming over the Sahara Desert and the recent recovery from the Sahelian drought. J. Climate, 28, 108123, https://doi.org/10.1175/JCLI-D-14-00039.1.

    • Search Google Scholar
    • Export Citation
  • Eyring, V., S. Bony, G. A. Meehl, C. A. Senior, B. Stevens, R. J. Stouffer, and K. E. Taylor, 2016: Overview of the Coupled Model Intercomparison Project phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev., 9, 19371958, https://doi.org/10.5194/gmd-9-1937-2016.

    • Search Google Scholar
    • Export Citation
  • Fasullo, J. T., 2010: Robust land–ocean contrasts in energy and water cycle feedbacks. J. Climate, 23, 46774693, https://doi.org/10.1175/2010JCLI3451.1.

    • Search Google Scholar
    • Export Citation
  • Folland, C. K., T. N. Palmer, and D. E. Parker, 1986: Sahel rainfall and worldwide sea temperatures, 1901–85. Nature, 320, 602607, https://doi.org/10.1038/320602a0.

    • Search Google Scholar
    • Export Citation
  • Girden, E. R., 1992: ANOVA: Repeated Measures. Sage Publications, 79 pp.

  • Hastie, T., R. Tibshirani, and J. Friedman, 2009: The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer, 745 pp.

  • Herrmann, S. M., A. Anyamba, and C. J. Tucker, 2005: Recent trends in vegetation dynamics in the African Sahel and their relationship to climate. Global Environ. Change, 15, 394404, https://doi.org/10.1016/j.gloenvcha.2005.08.004.

    • Search Google Scholar
    • Export Citation
  • Holton, J. R., 1973: An introduction to dynamic meteorology. Amer. J. Phys., 41, 752754, https://doi.org/10.1119/1.1987371.

  • Hu, J., L. Sun, C. H. Li, X. Wang, X. L. Jia, and Y. P. Cai, 2018: Water quality risk assessment for the Laoguanhe River of China using a stochastic simulation method. J. Environ. Inf., 31, 123136, https://doi.org/10.3808/jei.201800387.

    • Search Google Scholar
    • Export Citation
  • Huang, W. W., X. J. Chen, Y. R. Fan, and Y. P. Li, 2022: Management of drinking water source in rural communities under climate change. J. Environ. Inf., 39, 136151, https://doi.org/10.3808/jei.202000431.