Intensification of the hydrological cycle expected in West Africa over the 21st century

This study uses the high resolution outputs of the recent CORDEX-AFRICA climate projections to investigate the future changes in different aspects of the hydrological cycle over West Africa. Over the twenty-first century, temperatures in West Africa are expected to increase at a faster rate (+ 0.5 °C per decade) than the global 10 average (+ 0.3 °C per decade), and mean precipitation is expected to increase over the Guinea Coast (+ 0.03 mm/day per decade) but decrease over the Sahel (0.005 mm/day per decade). In addition, precipitation is expected to become more intense (+ 0.2 mm/day per decade) and less frequent (1.5 days per decade) over the entire West Africa as a results of increasing regional temperature (precipitation intensity increases on average by + 0.35 mm/day per °C and precipitation frequency decreases on average by – 2.2 days per °C). Over the Sahel, the average length of dry spells is 15 also expected to increase with temperature (+ 4% days per °C), which increases the likelihood for droughts with warming in this sub-region. Hence, the hydrological cycle is expected to increase throughout the twenty-first century over the entire West Africa, on average by + 11% per °C over the Sahel as a result of increasing precipitation intensity and lengthening of dry spells, and on average by + 3% per °C over the Guinea Coast as a result of increasing precipitation intensity only. 20


Introduction
It is now established that global warming will result from enhanced anthropogenic greenhouse gases (e.g. Collins et al. 2013). Such a warming is expected to affect precipitation and its variability, especially drought and flood episodes, in both the tropics and the subtropics (Zwiers et al., 2013;Giorgi et al., 2014). Over West Africa, previous studies (Collins 25 et al., 2013;Diedhiou et al., 2018;Bichet et al., 2019) have shown that the warming is expected to occur at a faster rate than the global average (+ 0.5 vs. + 0.3 °C per decade). Future changes in precipitation however are still unclear (e.g. Collins et al., 2013;Sylla et al., 2016;Bichet et al., submitted). Nevertheless, future changes in precipitation extremes are expected in some sub-regions, such as an increase in the maximum length of dry spells over West Sahel (Sylla et al., 2016;Diedhiou et al., 2018) and an intensification of extreme rainfall over the Guinea Coast .

30
Particularly relevant for agriculture, changes in precipitation are also projected during the growing season, expected to become shorter, as torrid, arid, and semi-arid climate conditions are expected to extend (Sylla et al., 2016). Such conditions can produce significant stresses on agricultural activities, water resources management, ecosystem services and urban areas planning over West Africa, a region that is already highly vulnerable to climate variability. However, whereas previous studies project important changes in the precipitation, very little is known about the role of future 35 warming and the processes involved.
Global distribution of tropospheric moisture and precipitation is highly complex, but there is one clear and strong control: moisture condensates out of supersaturated air. Assuming that relative humidity would remain roughly constant under global warming, the Clausius-Clapeyron relationship implies that specific humidity would increase exponentially with temperature, at a rate of about 6.5% per °C (e.g. Allen and Ingram, 2002). Assuming no change in the evapo-40 transpiration, a warmer atmosphere is thus expected to be able to hold more moisture before reaching saturation, thereby taking more time to reach saturation (longer periods of dryness between two rainy episodes), and releasing more water when moisture does condensate (intensification of the precipitation). Within this integrated view,  introduced a single index (HY-INT) that quantitatively combines measures of precipitation intensity and dry spell length, thereby providing an overall metric of hydroclimatic intensity.

45
To better understand the future impact of the warming on the hydrological cycle in the different sub-regions of West Africa, this study uses the state-of-the-art, high resolution projections of the recent CORDEX-AFRICA (Giorgi et al., 2009;Jones et al., 2011;Hewitson et al., 2012;Kim et al., 2014) experiments to investigate, over the twenty-first century, the future changes in different aspects of the hydrological cycle and their relationship with regional temperatures. After describing the methodology (Section 2), the expected changes in temperature, precipitation, 50 precipitation intensity, dry spells, wet spells, and HY-INT are identified (Section 3.1), before their relationship with regional temperature is quantified (Section 3.2). Section 4 discusses and concludes the study.

Methodology
We consider the three following sub-regions: West Sahel (10°N-20°N 18°W-10°W), Central Sahel (10°N-20°N 10°W-55 10°E), and Guinea Coast (5°N-10°N 10°W-10°E), shown as black boxes in Figure 1a. We focus on annual values over the period 2006-2099. Following previous studies (Froidurot et al., 2017;Bichet and Diedhiou, 2018a and 2018b), we define a wet (dry) day using the threshold of 1 mm/day. We define a dry spell as a sequence of 2 or more consecutive dry days, that are preceded and followed by a wet day. Hence, the duration of a dry spell, as defined in our study, spans from 2 to 365 days. We compute the annual precipitation intensity (INT), number of wet days (RR1), maximum length 60 3 of consecutive dry days (CDD), and maximum length of consecutive wet days (CWD) following the definition of the Expert Team of Climate Change Detection and Indices (ETCCDI; Zhang et al., 2011). Note that because INT corresponds to the precipitation averaged over wet days, a change in the INT value directly translates into a change in the intensity of wet events, regardless the number of wet events. In addition, we compute the annual contribution of very heavy rain (C98) following Eq. (1): Where is the sum of daily precipitation above or equals to the 98 th percentile annual value at wet days (Pctl98), and is the sum of daily precipitation at wet days during that year. Following previous studies Diedhiou, 2018a and2018b), we compute the annual average duration of dry spells (DSL) following Eq. (2): where is the annual number of dry days excluding isolated dry days (single dry day preceded and followed by a wet day), and is the total number of dry spells during that year. Note that the annual number of dry days is directly Where and are the normalized and , respectively. The normalization consists, for each grid point, in dividing the annual time series of by its mean value over the period 2006-2099.

Data
We use an ensemble of 18 high-resolution regional climate projections taken from the most up-to-date ensemble 80 produced in the recent years for Africa: CORDEX-AFRICA (Giorgi et al., 2009;Jones et al., 2011;Hewitson et al., 2012;Kim et al., 2014). All the simulations available online at the time of the analysis have been used. In this ensemble,  Furthermore, although averaging model output may lead to a loss of signal (such that the true expected change is very 4 likely to be larger than suggested by a model average), there is too little agreement on metrics to separate "good" and 90 "bad" models to objectively weight the models (Knutti et al., 2010). In the following, we thus use the equal-weighted model average to illustrate the mean response of our ensemble (multimodel mean maps in Figures 1-2 Dakar (Ouagadougou and Accra) than the observations. Nevertheless, we find that the observations are always included within the range of the 18 CORDEX simulations. Hence, we conclude that the CORDEX simulations compare satisfactorily well with the observations, and can be used for the purpose of our study.

125
Trends ( and an increase in the average length of dry spells (DSL). Hence over the Sahel, rainfall events are expected to become more intense and separated by much longer periods of dryness.

Relationship with temperature
Annual values ( Table 1 for color references), and the multimodel mean is represented by the thick black dots. According to Figure   3, mean precipitation decreases with temperature over the Sahel (multimodel mean decreases by -0.032 and -0.012 mm/day per °C over West Sahel and Central Sahel, respectively)  % of the changes in mean precipitation in all the three sub-regions, but more than 67 % (51 %) of the changes in RR1 165 (INT). As seen in Figure 3, even though the annual mean values vary greatly from a simulation to another (e.g. NCC-NorESM1-HIRHAM5 is particularly warm over Central Sahel, ICHEC-RACMO is particularly cold over the three subregions, NCC-NorESM1-HIRHAM5 is particularly wet over the Guinea Coast, and CSIRO-SMHI is particularly wet over the Sahel), the relation between each variable and the temperature is consistent across all models, albeit with a different strength.  Figure 4, the annual mean values of DSLn vary greatly from a simulation to 7 another over the Sahel, but are similar across simulations over the Guinea Coast. Similar to Figure 3, we find that the relation between each variable and the temperature is consistent across all models, albeit with different strength. We conclude that most of the trends observed in Figures 1 and 2 show a positive relationship with regional warming.

4 Discussion and conclusion
This study uses an ensemble of high resolution regional climate projections (CORDEX-AFRICA) to investigate, over the twenty-first century, the relationship between regional warming and different aspects of the hydrological cycle, as seen in three different sub-regions of West Africa. In agreement with previous studies (e.g. Vizy and Cook, 2012;Collins et al., 2013;Sylla et al., 2016;Diedhiou et al., 2018;Klutse et al., 2018), we find that 1) West

190
African surface temperatures are expected to increase at a faster rate than the global averaged warming (+ 0.5 °C vs. + 0.3 °C per decade), 2) precipitation is expected to intensify but rarefy over the entire region, 3) dry spells are expected to become longer (especially over the northern and the western part of the Sahel), and 4) wet spells are expected to become shorter over the Guinea Coast.
In addition, we show that 1) mean precipitation is expected to increase over the Guinea Coast and decrease 195 over the Sahel, and 2) the hydrological cycle, as defined by Giorgi et al. (2011), is expected to intensify over the entire West Africa (+ 5 % per decade on average). Whereas this intensification results solely from more intense precipitation over the Guinea Coast, we find it results from both, more intense precipitation (+ 2 %) and longer periods of dryness (+ 5-10 %) over the Sahel.
According to our results, all the aforementioned trends show a positive relationship with regional temperatures.

200
In agreement with Collins et al. (2013), we find that mean precipitation is expected to decrease with temperature over the Sahel and increase with temperature over the Guinea Coast. In addition, we find that the hydrological cycle is expected to increase with temperature over the entire West Africa, on average by + 11 % per °C over the Sahel and + 3 % per °C over the Guinea Coast. This increase is in qualitative agreement with the Clausius-Clapeyron relationship, which implies that specific humidity would increase exponentially with temperature (at a rate of about 6.5% per °C), 205 meaning that a warmer atmosphere is expected to take longer to reach saturation and release more water when it condensates, thereby intensifying the hydrological cycle (e.g. Allen and Ingram, 2002). Over the Sahel, we find indeed that the warmer atmosphere takes longer to reach saturation (DSL increases on average by + 7.5 % per °C) and releases more water when it condensates (INT increases on average by + 3.1 % per °C). Over the Guinea Coast however, we find that the warmer atmosphere does not take longer to reach saturation (DSL decreases on average by -1 % per °C), 210 but does release more water when it condensates (INT increases by + 4.1 % per °C but DSL decreases by -1 % per °C).
To understand the processes involved, Figure 5 (top row) shows the evolution of annual mean specific humidity as a 8 function of annual mean temperatures in all the three sub-regions. It is shown that specific humidity increases with temperature on average by + 5 % per °C in all the three sub-regions ( Figure 5), which is close to the rate expected from the Clausius-Clapeyron relationship (+ 6.5 % per °C). Thus, we conclude that in all the three sub-regions, a warmer 215 atmosphere does increase the amount of moisture in the atmosphere, which leads to more intense precipitation (INT increase in both sub-regions). However, whereas a warmer atmosphere also leads to longer periods of dryness over the Sahel (DSL increase over the Sahel), this is not the case over the Guinea Coast. We suggest that unlike the Sahel, the atmosphere over the Guinea Coast does not require more time to reach saturation because it is already very close to saturation. Thus, although the likelihood for droughts increases with temperature over the Sahel (and in particular over 220 West Sahel), this is not the case over the Guinea Coast. To understand the impact on the very heavy rainfall and floods, Figure 5 also shows the evolution of the 98 th percentile annual value (Pctl98 in mm/day, middle row) and the annual contribution of very heavy rain (C98 in %, bottom row) as a function of annual mean temperatures in all the three subregions. It is shown that a warmer climate implies heavier rainfall (multimodel mean increases by + 3.1 %, + 4.2 %, and + 8.6 % over West Sahel, Central Sahel, and Guinea Coast, respectively) and a larger contribution of very heavy rainfall 225 (+ 5.6 %, + 4.1 %, and + 3.7 % in West Sahel, Central Sahel, and Guinea Coast, respectively) in all the three subregions ( Figure 5), which indicates an increase in the likelihood for floods with temperature over the entire West Africa.
Finally, it is worth noting that over the Sahel, precipitation intensity is also driven by the frequency of the Mesoscale Convective Systems (MCSs), which are driven by the meridional temperature gradient between the Sahel and the Sahara (Taylor et al., 2017). Whereas the meridional temperature gradient between the Sahel and the Sahara is 230 projected to increase in CORDEX-AFRICA (on average by + 2 °C by the end of the century, not shown), the impact of this increase on the frequency of the MCSs cannot be simulated by these models (50 km) because MCSs occur on scales that are not resolved by these models. Hence, we suggest that over the Sahel, precipitation intensity may increase more than projected in our study, as result of the increasing meridional temperature gradient between the Sahel and the Sahara. For instance, Berthou et al. (2019) have shown that over the West Sahel, future changes in extreme rainfall 235 increase by a factor 5 to 10 at 4.5 km resolution (convection-permitting model allowing a good representation of MCSs), as compared to a factor 2 to 3 at 25 km resolution. Similarly, the impacts of atmospheric aerosols, particularly abundant over West Africa due to seasonal desert dusts (Konare et al., 2008;N'Datchoh et al., 2018), are only partially accounted for in CORDEX AFRICA due to the simplified parameterization schemes for aerosols in this dataset.
However, because aerosols are expected to affect temperature and precipitation in this region (e.g. Konare et al., 2008; 240 N'Datchoh et al., 2018), we suggest that our results are also limited by this simplified representation of aerosols.
Additional simulations at higher resolution and using a more complex parameterization scheme for aerosols would be required to identify the impact of MCSs and aerosols on our results, which is beyond the scope of our study.   Trends that are not significant at 95% according to the Student's t-test are shaded in gray.
255 Figure 3. Annual values of mean precipitation (mm/day), RR1 (days), and INT (mm/day) (y-axis) shown against annual mean temperature (°C) (x-axis), and averaged over a) West Sahel, b) Central Sahel, and c) Guinea Coast. Each color corresponds to a single simulation, as described in Table 1, and the thick black dots correspond to the multimodel mean.
Also shown are the fitted regression line of the multimodel mean (red line) and the associated coefficient of determination ('r 2 ') and correlation ('slope').  Table 1, and the thick black dots correspond to the multimodel mean. Also shown are the fitted regression line of the multimodel mean (red line) and the associated coefficient of determination ('r 2 ') and correlation ('slope').
265 Figure 5. Annual values of specific humidity, 98 th percentile (mm/day), and contribution of precipitation above the 98 th percentile (%) (y-axis)shown against annual mean temperature (°C) (x-axis), and averaged over a) West Sahel, b) Central Sahel, and c) Guinea Coast. Each color corresponds to a single simulation, as described in Program's Working Group on Regional Climate (http://www.cordex.org/data-access/esgf/) and the CHIRPS dataset from the Climate Hazards Group (http://chg.geog.ucsb.edu/data/chirps). Our work has benefited from access to rainfall data sets provided by the AMMA-CATCH observatory, the AMMA international program, DMN 290 Burkina, ANACIM, and DMN Niger; we sincerely thank all of them, as well as the staff at the IGE computation center (Guillaume Quantin, Véronique Chaffard, Patrick Juen, and Wajdi Nechba) for their technical support, and Geremy Panthou for his role in accessing the data and insights into the dataset. We also thank the staff at the IGE computation center (Patrick Juen and Wajdi Nechba) for their technical support.   Table 1, and the thick black dots correspond to the multimodel mean.
Also shown are the fitted regression line of the multimodel mean (red line) and the associated coefficient of determination ('r 2 ') and correlation ('slope'). 435 440