2.1 Data

The Climatic Research Unit (CRU) Time-Series (TS) v. 4.0 (^{Harris et al., 2014}) for the period 1901-2015 was downloaded in gridded form (0.5º × 0.5º) from https://crudata.uea.ac.uk/cru/ (accessed: June 18, 2017). The downloaded CRU TS4.0 data comprised minimum (T_min, ºC) and maximum (T_max, ºC) temperatures, as well as Pcp and PET (mm). Each variable was at a monthly temporal resolution.

The PET (mm day^-1) data of the CRU TS4.0 (^{Harris et al., 2014}) was computed using a variant of the Penman-Monteith method (^{Allen et al., 1994}) based on half degree gridded vapor pressure (hPa), wind speed (m s^-1), cloud cover (%), and absolute values of T_max, T_min, mean temperature (ºC), and wind speed (m s^-1). The monthly PET total (mm) used in this study is a sum of the daily PET values for each month.

The CRU datasets used in this study were obtained from gauge interpolations. Several areas across Africa (especially in sub-Saharan Africa) lack dense weather stations. Furthermore, several weather stations are not continuously operated due to poor maintenance of data recording or observation equipment coupled with interruptions by political turmoil (^{Onyutha, 2018b}). The accuracy of interpolation products depends on the number of gauge stations in a particular location. Therefore, the accuracy of results and the ranges of uncertainty vary across the continent.

Obtaining reliable estimates of evaporation (or flow of water from land to the atmosphere) remains a challenging task. Estimating evaporation using the distributed model requires a high resolution spatio-temporal description of land use and land cover types, and these spatial data are not always available. In some cases, estimates of terrestrial evaporation are made from satellites: however, the flux of terrestrial evaporation cannot be directly sensed from satellites (^{Martens et al., 2017}). Nevertheless, advances have been made in the improvement of evaporation estimates over large scales. Some examples include data-driven estimates of global land evapotranspiration from 1982 to 2008 (^{Jung et al., 2010}), understanding the response of terrestrial evaporation to teleconnection patterns ^{(Martens et al., 2018}), and the Global Land Evaporation Amsterdam Model (GLEAM) datasets (^{Miralles et al., 2011}; ^{Martens et al., 2017}), downloadable from https://www.gleam.eu/ (accessed: June 14, 2020); however, evaporation datasets based on satellite information are normally comprised by short-term records. The CRU TS4.0 datasets were adopted for this study because of their reliability and long-term records gridded at a spatial (0.5º × 0.5º) scale covering the entire African continent. Since the number of weather stations influences interpolation products at a given location, spatial similarities of results from analyses based on the CRU TS4.0 data should not be assessed at the level of individual locations, but over a region or large area (^{Onyutha, 2018a}).

The data at grid points over land were thoroughly checked and confirmed to have no missing values. The various countries within the continent can be seen in Figure SM-1 of the supplementary material. The long-term mean of the data used in this study can be seen in Figures SM-2 to SM-6 of the supplementary material.

2.2 Trend analyses

To consider the influence of seasonality (here, a season is taken as a month) on the results of trend analyses, two steps should be taken: (1) the trend statistic values are computed separately for each month, which is done for both trend directions and magnitudes, and (2) trend results for the various months are combined into an overall or global trend measure.

2.3. Variability analyses

There are two ways in which variability can be quantified to test H ₀ of natural randomness using the CSD method, namely the use of variability CSD statistic and temporal variation of CSD-based subtrends.

3.1 Trends and variability

3.1.1 Trends

Time series plots for Pcp and PET (Fig. 1a-e) were made using data extracted at a number of locations from different climatic conditions in the continent, including warm semi-arid climate (WSAC), warm desert climate (WDC), equatorial climate (EC), humid sub-tropical climate (HSTC), and cold semi-arid climate (CSAC). The Pcp average from the different climate types can be seen in Figure 1f. The location (longitude and latitude) at which the data for WSAC, WDC, EC, HSTC, and CSAC were extracted was (-11.25º, 13.25º), (1.25º, 24.25º), (20.25º, 0.25º), (24.75º, -15.75º), and (25.75º, -27.75º), respectively. What is noticeably common for the data of different climate types is the positive trend in PET (Fig. 1a-e) from around 1985 to 2015 (end of data). Furthermore, the decrease in Pcp (Fig. 1a-d) from around 1960 to the end of the data period is also generally evident across the different climate types. At some locations (see Fig. 1c, e) series before 1940 showed minimal deviations from a certain constant value, which could be due to a data quality problem. It is possible that such locations had only a few weather stations before 1940. For the desert climate (Fig. 1b), annual Pcp was generally lower than for other types of climate.

Fig. 1 Temporal variation of annual Pcp and PET for: (a) warm semi aridclimate, (b) warm desert climate, (c) equatorial climate, (d) humid subtropical climate, and (e) cold semiarid climate, (f) average of (a) to (c) data. 

Trends results in each month’s data can be observed in Figures SM-7 to SM-14 of the supplementary material. Figure 2 was obtained based on the trends results of the various months. In general, the entire African continent (except Madagascar) was characterized by warming trends with respect to T_min and T_max. Large increases in both T_max and T_min at rates between 0.08 and 0.25 ˚C per decade were observed in Sudan as well as the northwestern and southern parts of Africa (Fig. 2e-f). For these increases of T_max and T_min, H ₀ of no trend was rejected (p < 0.01 or |Z| > 2.57) (Fig. 2a-b) in most locations. For the cooling trends in Madagascar (Fig. 2d-e), H ₀ was not rejected (p > 0.05 or |Z| < 1.96) (Fig. 2a-b). Except over some West African countries, the spatial distributions of Z values for T_max and T_min were comparable (Fig. 2a-b).

Fig. 2 Overall standardized trend statistic Z for monthly (a) T_max (ºC), (b) T_min (ºC), (c) PET (mm), and (c) precipitation (Pcp, mm), and trend slopes for (e) T_max, (f) T_min , (g) PET, and (h) Pcp, averaged for all months from January to December. 

Most parts of the continent were characterized by monthly PET increases at rates between 0.1 and 0.04 mm year^-1 (Fig. 2g). Large increases in PET at rates between 0.05 and 0.085 mm year^-1 were observed along the western coastline of South Africa, northwestern areas like Tunisia, as well as some areas around Lake Victoria in the equatorial region of East Africa. For these increases, H ₀ was rejected (p < 0.01 or |Z| > 2.57), as seen in Figure 2c. Along the coastline of the Gulf of Guinea, patches in some areas exhibited decreases in PET at rates as low as -0.077 mm year^-1. However, for these decreases, H ₀ was not rejected (p > 0.01 or |Z| < 2.57), as seen in Figure 2c.

Monthly Pcp across the continent was mainly characterized by drying at rates as low as -0.20 mm year^-1 (Fig. 2h). Pcp decreases at rates between -0.3 and -0.2 mm year^-1 were confined to the coastal areas of Guinea, Sierra Leone, and Gambia (Fig. 2 f). However, for the decreases, H ₀ was not rejected (p > 0.05 or |Z| < 1.96). Furthermore, Pcp increases were mainly of magnitudes between 0 and 0.1 mm year^-1. These increases were shown in the Great Lakes region, southwestern Africa and Madagascar (except its southernmost area) (Fig. 2h). However, for both Pcp increases and decreases, H ₀ was mostly not rejected (p > 0.05) (Fig. 2d).

3.1.2 Variability

Figure 3 shows an overall average of the CSD variability Z* values obtained from data of various months (as can be seen from Figs. SM-15 to SM-18 of the supplementary material). The values 1.64, 1.96, 2.58, and 2.81 in the legend of Figure 3 are the thresholds for rejecting H ₀ of natural randomness at α = 0.10, 0.05, 0.01, and 0.005, respectively. Areas with low variability (Z* > 2.81) in T_max (T_min) were mainly between 10º S and 5º N (10º N and 10º S). High variability (Z* < 1.63) in temperature was mainly found in Mozambique, Egypt, Mauritania, Western Sahara and around Lake Victoria (Fig. 3a-c). For Pcp, high variability was exhibited in South Africa, West Africa (especially along the Gulf of Guinea), East Africa, Madagascar, and areas with warm Mediterranean climate (Fig. 3d). Pcp variability was low (Z* > 2.81) across the Sahara and Namib-Kalahari deserts. The significance of the variability statistic was found to vary from one region to another (Fig. 3a-d), which based on Figs. SM-15 to SM-18 of the supplementary material was found to be dependent on the month being considered.

Fig. 3 Overall standardized variability statistic Z* for (a) T_max, (b) T_min, (c) PET, and (d) Pcp based on data from January to December. 

3.2 Covariability of subtrends from the selected climatic variables

Results for the correlation between Pcp subtrends and those of T_max, T_min and PET can be seen in Figures 4, 5, and 6, respectively, in which the values 0.25 and 0.31 are the thresholds for rejecting H ₀ of no correlation at α = 0.01 and 0.001, respectively. From May to September, a strong negative correlation between Pcp and T_max (Fig. 4e-i), PET (Fig. 6e-i) and T_min (Fig. 5e-i) was confined to the region between 4-21º N. In southern Africa (generally south of 8º S), a strong negative correlation between Pcp and T_max was exhibited from January to March, May, October and December (Fig. 4a-c, e, j, l). A strong negative correlation (p < 0.001) between Pcp andT_max or Pcp and PET was observed in the February, March, October, and November data, especially along the coast of the Mediterranean Sea. A strong positive correlation between Pcp and T_min of March (Fig. 5c) was observed in Mozambique, Zambia and Zimbabwe. However, in October (Fig. 5j) and August (Fig. 5h), the strong positive correlation between Pcp and T_min was mainly confined to South Africa and the coastal areas along the Gulf of Guinea, respectively. A positive correlation between Pcp and T_max (Fig. 4g-h) as well as Pcp and T_min (Fig. 5g-h) during July and August was observed along the Gulf of Guinea, as well as the north of Morocco and Algeria. Over East Africa, a strong negative correlation (p < 0.001) between Pcp and T_max was osberved in April and November (Fig. 4d, k). In general, in January and February, areas with a strong correlation between Pcp and temperature or PET were far south of the equator. From March through April and May to June, in areas where H ₀ was rejected (p < 0.001) for a strong negative correlation the spatial coherence decreased consistently, but with increasing shifts northwards as far as 14º N. Slightly past the middle of the year (July or August), a band of strong negative correlation was found in the Northern Hemisphere as far as 20º N. The described shifts in areas with strong correlation between Pcp and temperature were more evident for T_max than for T_min.

Fig. 4 Correlation coefficients between Pcp and T_max for (a) January to (l) December. 

Fig. 5 Correlation coefficients between Pcp and T_min for (a) January to (l) December. 

Fig. 6 Correlation coefficients between Pcp and PET for (a) January to (l) December. 

Generally, it is noticeable that in most areas, correlation coefficients between subtrends of Pcp and T_max (Fig. 4a-l), Pcp and T_min (Fig. 5a-l), and Pcp and PET (Fig. 6a-l) were mainly negative. Evidently, these negative correlations coefficients were mostly weak (p > 0.01) or fell in the range between 0 and -0.24. However, areas with strong negative correlations were larger for T_max and PET than that for T_min. On the other hand, areas with positive correlation were larger for T_min than for T_max and PET. Furthermore, by comparing maps for corresponding months, correlation coefficients between Pcp and T_max (Fig. 4a-l) were more comparable spatially with the results for PET (Fig. 6a-l) than for T_min (Fig. 5a-l).

Figure 7 shows the results of wavelet coherence analysis. The reliable data within the “cone of influence” away from the edge effects are indicated by the colored region. Significantly coherent areas are enclosed in bold black lines derived with Monte Carlo randomizations. The strength of coherence between the time series at each period through time can be seen in terms of colors. The color at the bottom (top) of the legend bar indicates low (high) coherency.

Fig. 7 Wavelet coherence plots for (a) Pcp and PET, (b) Pcp and T_max, (c) Pcp and T_min, (d) PET and T_max, (e) PET and T_min, and (f) T_max and T_min. 

The strength of the Pcp with T_max, T_min or PET covariation was inconsistent through time (Fig. 7a-c). The strongest coherence between Pcp and PET occurred at multiple scales (2-5 years, Fig. 7a). Coherence between Pcp and PET was strong over the periods 1905-1925 and 1985-2005 (Fig. 7a). Pcp cycled in anti-phase with PET from 1985 to 2005. The strongest coherence between Pcp and T_min (or T_max) occurred at multiple scales (6-8 years, Fig. 7b-c). Strong coherence of Pcp with T_max and T_min was observed over the periods 1901-1935 (Fig. 7b) and 1901-1925 (Fig. 7c), respectively. Pcp cycled in anti-phase with T_max from 1901 to 1935. Coherence between PET and temperature remained strong at multiple scales (from 1 to about 30 years) over the entire data period (1901-2015). Arrows in this figure show phase relationships between the variables within the areas of strong coherence. In Figure 7d-f, the arrows point to the right meaning that PET and T_max, PET and T_min, as well as T_min and T_max are in phase. This indicates that PET has a strong positive influence from temperature. Some arrows in Figure 7a-c point to the left indicating that Pcp and PET or Pcp and temperature are in anti-phase. From 1901 to 1935, Pcp cycled downwards with T_min at multiple scales (15-16 years), which means that T_min led Pcp by π × 2-1. However, Pcp cycled upwards with T_min at multiple scales (1.4 years, Fig. 7 c). This means that T_min led Pcp by π × 2-1.

4.1. Trends and variabilty results

The results of this study are consistent with the findings of several previous studies, especially regarding trends. The confidence level in the evidence of warming over land regions across the African continent is high (^{Niang et al., 2014}). Throughout the 20th century, the African continent warmed at a rate of 0.58 ºC per decade (^{Hulme et al., 2001}). A significant rise in African temperature occurred between 1979 and 2010 (^{Collins, 2011}).

Regarding the equatorial region (specifically around Lake Victoria in East Africa), ^{Nyeko-Ogiramoi et al. (2013)} and ^{Kizza et al. (2009)} found mainly positive trends in the seasonal rainfall. Furthermore, T_min and T_max over the period 1970-2010 generally exhibited trends to rising temperatures (Nyeko-Ogiramoi et al., 2013). Significant warming and drying were confirmed in the western equatorial Africa (^{Bush et al., 2020}). Pcp and T_min data from the weather of Lopé in Gabon, observed over the period 1984-2018, revealed that warming occurred at a rate of +0.25 ºC per decade and this was accompanied by a drying rate of −75 mm per decade (^{Bush et al., 2020}).

The West African Sahel was characterized by a general warming trend from 1960 to 2010 (^{Ly et al., 2013}). Even the data from various stations across the Volta basin of West Africa between 1960 and 2002 exhibited significant warming trends (^{Neumann et al., 2007}). Pcp exhibited a decrease, especially along the Gulf of Guinea and the Congo River basin (^{Onyutha, 2018a}).

For Egypt and Sudan, this study showed that: (i) T_max and T_min exhibited increasing trends at rates from 0.041 to 0.25 ºC per decade, (ii) Pcp decreased at rates between -25 and 0 mm per decade, and (iii) PET yielded positive trends with magnitudes on the range from 1.32 to 4.8 mm per decade. ^{El Kenawy et al. (2019)} found that T_max in Egypt exhibited a stronger warming than T_min in the period 1983-2015. Conversely, ^{Nashwan et al. (2019)} showed that T_min increased much faster (0.08-0.29 ºC per decade) than T_max (0.07-0.24 ºC per decade) across Egypt. Rainfall in Egypt was also characterized by decreasing trends (^{Gado et al., 2019}).

For South Africa, ^{Karl et al. (1993)} also reported positive trends for both T_max and T_min over the period 1951-1991. According to ^{MacKellar et al. (2014)}, T_max in South Africa from 1960 to 2010 exhibited a positive trend. However, decreasing trends in both T_max and T_min between 1940 and 1989 were reported in South Africa by ^{Muhlenbrunch-Tegen (1992)}. Furthermore, ^{Hulme et al. (2001)} showed a cooling tendency over the coastal areas of South Africa considering the period 1901-1995. In this study, cooling was not evident in the coastal areas of South Africa, in contrast with Madagascar (near the eastern coastal area of South Africa). Some possible discrepancies of the results in this paper with those of previous studies are due to differences in the data periods used, for instance, 1901-1995 (^{Hulme et al., 2001}), 1940-1989 (^{Muhlenbrunch-Tegen, 1992}), and 1901-2015 in the present study. Trend results depend on the period considered for analyses (^{Onyutha et al., 2016}). Another reason is the effect of seasonality, which was explicitly considered in this study but ignored by previous researchers.

4.2 Covariation of Pcp with T _max , T _min , and PET

The general negative correlation coefficients between Pcp and temperature subtrends indicate that as temperature increases, total Pcp decreases and vice versa; however, as temperature increases, PET also increases. In other words, a direct relationship exists between temperature and PET. High temperature implies an increase in the amount of energy for the conversion of liquid water to water vapor. Also, an increase in temperature widens plant leaves’ stomata, allowing more water vapor to escape than under conditions with reduced temperature. Water demands to meet evaporation and transpiration are supplied by Pcp. Thus, as Pcp increases, PET is also expected to increase. However, PET rates are controlled by several factors besides Pcp, such as: (i) humidity (evaporation rates decrease when an area is too humid), (ii) wind speed (evaporation rates increase when air is moving or humidity is being cleared by the wind), (iii) soil moisture (in dry soil, nature permits infiltration and sufficient water uptake by plants or vegetation prior to evaporation), and (iv) vegetation or plant type (while trees and crops are generous in losing water, some plants like cacti naturally prefer holding onto their water instead of easily giving it up for transpiration). Importantly, dry conditions mean more sunshine and less evaporative cooling, especially over land.

At some locations where H ₀ was rejected (p < 0.001), multi-decadal temporal changes in temperature could be inferred from the variations in Pcp. However, considering the long-term, it is evident that the temperature-Pcp covariation depends on the period used for analysis, which also applies to the relationship between PET and Pcp. For instance, as shown in Figure SM-2 of the supplementary material, T_max, PET and T_min generally exhibited negative trend slopes in the period 1941-1978; however, positive trends were observed in the periods 1901-1940 and 1979-2015. On the other hand, Pcp was on average characterized by positive and negative trends over the periods 1901-1960 and 1961-1990, respectively. Thus, trend slopes in both temperature and Pcp, as well as PET, were positive from 1901 to 1940; however, from 1979 to 1990, temperature and PET increased while Pcp declined over time. In fact, a rise in temperature increases the atmospheric evaporative demand and eventually Pcp is expected to increase. This is because high temperatures enhance the increase of warm moisture, thereby producing convection. Despite the general warming across Africa, some regions were characterized by drying trends, while in other areas Pcp increased. Besides, the warming trends across various regions of the continent are due to different influences. For instance, tropical North Africa is very warm due to a number of reasons, including: (i) because of its low latitude it receives a large amount of incoming solar radiation, (ii) its elevation is mostly less than 1000 masl, and (iii) advection of cool oceanic air from the tropical North Africa towards the interior (Sahelian and Sudanian belts) is inhibited by the general easterly trade winds coupled with the size of the African continent north of the equator (^{Moron et al., 2016}).

Pcp totals differ across the various regions of the continent, which can be thought of in terms of the influence from the shifts in ocean currents and wind patterns. For instance, heat motion is majorly influenced by ocean dynamics while the cycles of the El Niño Southern Oscillation (ENSO) responsible for the wet and dry conditions in several areas follow atmosphere-ocean interactions. According to ^{Trenberth et al. (2010)}, during the La Niña phase there is a major uptake of heat by the ocean. The absorbed heat is stored in the tropical western Pacific. The ocean cools as the atmosphere responds with typical El Niño weather patterns forced from the region (^{Trenberth et al., 2010}), which tends to be a typical source of influence on Pcp across various areas of the world.

The monthly spatial transition in the temperature-Pcp correlation (described in the results section) is typical of the latitudinal migration of the Inter-Tropical Convergence Zone (ITCZ). However, the analogy with the ITCZ migration was clearer for T_max than for T_min. This is because the sensitivity of temperature extremes to local influences is higher for T_min than for T_max. Pcp patterns and distributions across Africa dominantly rely on the migration of the ITCZ. A band of Pcp moves with the ITCZ as it migrates from its southernmost location (in January) to the northern extreme (in July). The lack of a high spatial coherence in the temperature-Pcp correlation over the equatorial region is due to the joint impacts of the Great Lakes and the high mountains on Pcp and temperature. The joint impact of the lakes and upslope breezes has considerable influence on convection, and subsequently on temperature. It is also worth noting that in cool highland areas, Pcp tends to be associated with above normal temperature; however, in the lowlands and hot environments, as well as the onset of the rainy season, Pcp occurrence is associated with below normal temperature (^{Camberlin, 2017}).

4.3. General implications of the warming and drying trends

The warming trends found in most parts of the continent have wide-ranging implications. A positive trend in temperature means an increase in the intensity of heat waves, thereby leading to many cases of illnesses or even deaths in several regions of the continent, given the high vulnerability of local populations. Continued increases in temperature present high possibilities of severe and/or persistent dry conditions. This is because PET depends on T_max and T_min. Thus, positive trends in T_max and T_min imply an increase in total PET, which in turn leads to a decline in soil moisture (especially if Pcp is insufficient) and severe lack of water for crops. By inducing water and heat stress, dry conditions (due to soaring temperatures, high evaporation, and subsequent decline in soil moiture) negatively affect crop production. Increasing temperature and declining Pcp totals lead to crop failure, thereby exacerbating food insecurity, which is already a formidable challenge to the subsistence of poor populations across Africa (especially in the sub-Saharan region). The dry and wet conditions in several locations across the continents tend to be characterized by late onset of Pcp, reduced length of the wet season (or longer than normal dry conditions), and soaring temperatures. Such changes tend to greatly affect dryland ecosystems, which are sensitive to changes in temperature and Pcp.

To meet the requirements of predictive adaptation in areas or regions with significant positive trends in temperature (T_max and T_min) and PET, smallholder farming should be based on drought-tolerant crop varieties. Good farming practices (for instance, using mulching to improve soil moisture and fertility) should be encouraged in smallholder farming; however, tackling food insecurity amid uncertain climatic conditions, especially in sub-Saharan Africa, requires an interplay of science and policy (^{Onyutha, 2019}). For instance, policy makers should ensure quick variety testing, as well a fast approval of the new crop varieties being developed. This should be done while addressing factors that hamper the adoption of scientifically improved crop varieties by farmers. Global collaboration regarding scientific research that supports agriculture is crucial to tackle food insecurity in Africa. Further measures should be taken, including increasing farmer’s access to markets, promoting non-farming activities, minimizing the differences in priorities across various subsectors of farming in sub-Saharan Africa, and addressing disparity in initiatives of regional and national dynamics (Onyutha, 2018c).

Deviations of Pcp and temperature from their normal conditions increase the risk of conflicts (^{Hsiang and Meng, 2014}; ^{Hendrix and Salehyan, 2012}). The way in which local populations respond to the impacts of increased warming and drying may lead to various other repercussions. Migration as an adaptive response to local environmental pressures (^{Tacoli, 2009}; ^{Warner, 2010}) leads to health risks (^{Serdeczny et al., 2016}) and tensions between ethnic groups, political and legal restrictions, and competition for (and limitations on) access to land (Tacoli, 2009). According to ^{Burke et al. (2009)}, the likelihood of civil war is higher in hotter years. In other words, warming increases the risks of civil wars in Africa (^{Burke et al., 2009}). Conversely, ^{Aldhous (2010)} remarked that civil war in Africa has no link to global warming. Thus, climate should not be blamed for African civil wars (^{Buhaug, 2010}). Nonetheless, the exacerbation of conflicts by the influence of drying and warming trends cannot be disregarded depending on a number of factors considered for analysis, such as: (i) which indicators of conflicts are used, (ii) which region is considered, and (iii) which period (or time frame) is chosen. Besides, drying and warming trends may lead to conflicts indirectly, thus the origins will be difficult to unearth. This could be the reason for a lack of evidence on the relationship between violent conflicts and climate change (^{Gleditsch, 2012}). It remains a fact that local populations will always tend to adapt to prolonged local environmental stresses. For instance, due to prolonged conditions of extreme heat and severe drought, pastoral communities adapt by moving from one place to another. This migration may be further and wider than in short-lived conditions of heat and water stress. Competition for resources (such as water, land, and pasture) by residents and immigrants may heighten conflicts. In order to alleviate poverty in local communities striving to adapt to the impacts of warming and drying trends on livelihood, policies should be established to promote non-farming incomes and creating jobs and employment opportunities for the rural population.

Increasing temperature and PET, alongside declining Pcp totals, imply a decrease in rainfall-runoff volumes. Thus, if warming and drying trends continue, the future water supply from rivers will be characterized by seasonal shortages. There are several applications which depend on volumes of water in the rivers, some of which include irrigation, reservoir operations, hydropower scheduling, and flow control for ecological purposes. Increasing water demand for irrigation and hydropower production is triggered by pressure arising from high population growth, especially in the sub-Saharan region. It is known that groundwater is the key source of safe drinking water in that region (^{MacDonald et al., 2009}). However, drying and warming trends also pose a serious risk for the reduction of volumes of water to recharge groundwater. In adapting to the impacts of warming and drying trends there should be a sustainable use of water resources.

4.4. Implications of changes in PET, temperature and Pcp for arid environments

The way in which changes in temperature, Pcp and other climatic factors affect regional evapotranspiration and net primary productivity in various arid environments of the world has been widely studied (^{McVicar et al., 2011}; ^{Bai et al., 2014}; ^{Onyutha, 2016c}). Variations in Pcp were found to significantly (p < 0.05) explain changes in PET across the River Nile riparian countries (^{Onyutha, 2016c}). In another relevant study for an arid environment (though in the northwest of China), ^{Bai et al. (2014)} quantified the contributions of agricultural oasis expansion, management practices and climate change to net primary production and evapotranspiration. Furthermore, ^{McVicar et al. (2011)} assessed the impact of trends in the observed terrestrial near-surface wind speeds for evaporation.

The results of this study are vital for an insight into the ecohydrology of water-limited ecosystems. In arid regions, evapotranspiration comprises a significant water budget component (^{Glenn et al., 2010}). Generally, water (followed by nitrogen) is the most vital limiting variable for plant growth in arid conditions (^{Smith et al., 1997}). Therefore, in arid regions, the plant community composition, function, and structure tend to be greatly influenced by possible controls on water movement (^{Loik et al., 2004}). Eventually, the amount of water that becomes available for root water uptake is also affected (^{Young et al., 2009}). On one hand, it may be claimed that water can be efficiently used by vegetation in dryland ecosystems (^{Huxman et al., 2004}); on the other hand, water use efficiency might vary based on the composition of plant species over the landscape (^{Huxman et al., 2005}). As opposed to arid regions, semi-arid systems tend to be mainly driven by Pcp. Long-term changes in climatic variables such as PET and Pcp may be linked to the impacts of human factors. For instance, the rapid expansion of cropland and intensive agricultural management practices can greatly affect regional carbon and water budgets (^{Bai et al., 2014}). Other relevant human factors which lead to changes in climatic variables include over-grazing, massive deforestation, transition in land-use or cover, and urbanization. An insight into the present and potential future vegetation and water availability in arid environments may be obtained by understanding how the ecosystem evapotranspiration is regulated (^{Wilske et al., 2009}).

Positive trends in PET (and temperature) imply an increase in the demand of crop water requirements. The cropping system in sub-Saharan Africa is almost entirely rainfed. However, addressing the water shortage issue to enhance crop yield implies an increase in irrigation requirements, but in irrigated and desert areas water shortage in turn indirectly affects soil salinity of the root zone (^{Abderrahman et al., 1991}). Especially under elevated water requirements, plants tend to consume some water from the leaching fraction thereby increasing soil salinity and causing a subsequent yield loss (^{Abderrahman et al., 1991}).

Results of long-term trends and decadal changes in evapotranspiration (and Pcp) indicate whether a region will be wetter or drier than its past climatic conditions. The findings of this study can be useful for a careful planning of water resources management, for instance, to determine how much water can be used to support environmental needs after safely being allocated for human use (^{Bunting et al., 2014}). Long-term increase of PET and a negative trend in Pcp imply that the PET to Pcp ratio increases over time. The evapotranspiration to Pcp ratio is important in arid and semiarid ecosystems (^{Glenn et al., 2014}) because water that is not lost through evapotranspiration can produce runoff and groundwater recharge, thereby influencing erosion, aquifer properties and regional stream flows (^{Milly, 1994}). Furthermore, changes in ecohydrological variables such as the PET to Pcp ratio can be taken as indicators of the transition in vegetation communities, especially across drylands (^{Heilman et al., 2014}). For instance, the evapotranspiration to Pcp ratio varies across woodland, grassland and shrubland ecosystems.

Before 2000, the Sahara Desert was shown to be changing in size by both expanding and contracting at some points over time (^{Tucker et al., 1991}; ^{Tucker and Nicholson, 1999}). Recently, an amplified warming trend has also been found in the Sahara Desert (^{Cook and Vizy, 2015}; ^{Zhou, 2016}). The results of this study, which indicate an increasing trend in long-term (1901-2015) temperature and PET, accompanied by Pcp decrease across North Africa, point towards an expansion of the Sahara Desert, whose variation in size (especially the increase) has major implications in terms of the influence of Saharan dust on sea surface temperatures as well as the occurrences of Atlantic hurricanes (see ^{Evan et al., 2016}). If the Sahara Desert continues to expand it will have significant socioeconomic and environmental impacts. For instance, the affected population will have to live under abnormal water-stressed conditions. Furthermore, many people will be displaced by harsh environmental conditions. The question should be: What steps need to be taken? It is worth noting that apart from low rainfall and dry conditions, other factors which enhance desertification and subsequent degradation of soils include dust transport and wind erosion (^{McLeod, 1976}). Therefore, some of the critical steps to be followed in line with desertification include reinstating degraded soil ecosystems and scaling-up land use and/or land management practices which are sustainable.