Abstract

The study presented here makes use of about 300 daily rain gauges covering a 1 700 000 km2 area in order to characterize the rainfall regimes of West Africa at hydrological scales. The rainfall regime is analyzed as a combination of two variables, the average number of events over a given period of time (nT) and the average cumulative rainfall per event (h). These two parameters are a measure of the occurrence rate and magnitude of the convective storms that generate most of the rainfall in this region. They define the average water input to the hydrological systems and the average time available for this water to be redistributed into the continental hydrological cycle before a new input occurs. By analyzing for a period of 40 yr (1951–90), the space and time variations of these two parameters, it is possible to better understand how the intraseasonal to decadal rainfall variability may impact on the hydrological cycle. The analysis is carried out in two steps. First, the annual cycle and migrations of the weather zones characterizing the climate of West Africa are considered. This leads to evidence of a sudden and synchronous rain onset between 9° and 13°N, which does not follow the classic scheme of a progressive migration of the rain zones, north and south with the sun. Second, the differences in the rainfall regimes between the two succeeding subperiods of 20 yr are obtained, the subperiod P1 (1951–70) being wet and the subperiod P2 (1971–90) being dry. The difference—averaged over the 16° by 12° study region—of the mean interannual rainfall between the wet and the dry periods is 180 mm yr−1. This difference is relatively evenly distributed in space, with no clear meridional gradient. Between these two periods, the parameter nT displays a systematic decrease, which appears well correlated to the decrease of the mean interannual rainfall. The variations of h are, by contrast, smaller in amplitude and more erratically distributed in space. When looking at the intraseasonal scale, it appears that the rainfall deficit of the dry period is primarily linked to a deficit of the number of events occurring during the core of the rainy season over the Sahel, and during the first rainy season for the region extending south to 9°–10°N. It is also shown that, in the south, the dry period is characterized by a shift in time of the second rainy season. All these characteristics have strong implications in term of agricultural and water resources management. They also raise questions about the traditional scheme used to characterize the dynamics of the West African monsoon.

1. Introduction

The semiarid regions of the world are known for their unreliable rainfall, which has a large impact on the continental hydrological cycle, the water resources and the food security. The Sahel, extending across Africa from the Atlantic Ocean to the Indian Ocean is the largest of these regions. The famines that struck the Sahel in the 1970s (1972–74) and in the 1980s (1983–85) have prompted a number of authors (e.g., Folland et al. 1986; Fontaine and Janicot 1996; Hastenrath 1990; Lamb 1983; Lamb and Peppler 1992; Nicholson 1981) to investigate possible mechanisms responsible for these dramatic events. In fact, these two sequences of a few extremely dry years are part of a longer drought that lasted from the end of the 1960s to the mid-1990s (Le Barbé and Lebel 1997a; D'Amato and Lebel 1998). This is well illustrated by Fig. 1, which shows a normalized rainfall index computed over the whole Sahel from a few long stations in Fig. 1a and from a denser network for the last 50 years over southern Niger in Fig. 1b.

Fig. 1.

(a) Evolution of the standardized rainfall index over the Sahel between 1921 and 1994; (b) evolution of the standardized rainfall index over Niger (south to 15°N) between 1950 and 1997. (The standardized rainfall index is an average of rainfall standardized at each station with respect to the mean of the station: R* = (Rm)/s where m and s are the mean and standard deviation of the rain series at the station.)

Fig. 1.

(a) Evolution of the standardized rainfall index over the Sahel between 1921 and 1994; (b) evolution of the standardized rainfall index over Niger (south to 15°N) between 1950 and 1997. (The standardized rainfall index is an average of rainfall standardized at each station with respect to the mean of the station: R* = (Rm)/s where m and s are the mean and standard deviation of the rain series at the station.)

This unusual dry spell was not limited to the Sahel, defined as the region extending north to 12°N, but extended to regions more to the south as well (see, e.g., Le Barbé and Lebel 1997b). This is clear from Fig. 2. The average rainfall deficit of the 1970s and 1980s with respect to the 1950s and 1960s is about 180 mm for the area covered by this figure, with a fairly regular distribution in space. An illustration of the effect of the drought in the more humid parts of West Africa is the shortage of electricity that struck the large coastal capitals, like Abidjan and Cotonou, during the summers of 1984 and 1998, following the dryness of the preceding rainy seasons. The low rainfall caused the reservoirs to remain empty and the hydropower plants had to be shut down. Such a large-scale phenomenon goes beyond the rainfall variability that commonly characterizes tropical rainfall regimes.

Fig. 2.

Map of the absolute rainfall differences between the wet period P1 (1951–70) and the dry period P2 (1971–90) over the region of study. Isolines are in mm yr−1

Fig. 2.

Map of the absolute rainfall differences between the wet period P1 (1951–70) and the dry period P2 (1971–90) over the region of study. Isolines are in mm yr−1

Recent diagnostic studies (Fontaine et al. 1998; Janicot et al. 1996; Ward 1998) have provided some interesting elements showing that the forcing of the West African rainfall by worldwide SSTs might have been modified during the 1960s. The continental surface conditions also play a role as shown by modeling studies (Polcher 1995; Eltahir and Gong 1996; Zheng and Eltahir 1998) and observations (Taylor and Lebel 1998). Diagnostic studies generally consider rainfall patterns at large spatial scales. This is because they either aim at developing tools for seasonal prediction or are limited by the resolution of the models. From a hydrological point of view, it is essential, on the other hand, to consider the scales of the processes influencing the partition of rainwater into the various components of the water cycle. This is also true when considering timescales.

The study presented here makes use of a set of 300 daily rain gauges covering a 1 700 000 km2 area (14° × 10°) in order to characterize the rainfall regimes at the event scale and their modification when comparing the wet and the dry periods mentioned above. It also provides some interesting clues to the dynamics of the rainy season(s) over West Africa.

2. Region of study and data used

In West Africa the datasets used in diagnostic studies are usually 10-day or monthly accumulations. The corresponding daily observations are not as readily available, even though there has been a sufficient number of rain gauges in operation since the midcentury to get a good picture of the spatial pattern of the rainfall regimes for timescales less than 10 days. In French speaking countries of West Africa, daily rain data have been regularly collected by the national meteorological services and delivered to a database built by the Centre Inter-Etats d'Etudes Hydrauliques (CIEH) and the Office de la Recherche Scientifique et Technique Outre-Mer (ORSTOM). The CIEH–ORSTOM database contains data until 1984. This database was updated to 1990, with the help of AGRHYMET—the technical centre of the Comité permanent Inter-états de Lutte contre la Sécheresse au Sahel—for three Sahelian countries (Burkina Faso, Mali, Niger) and with the help of the respective meteorological services for six other countries (Benin, Ivory Coast, Ghana, Guinea, Nigeria, and Togo). In the database that was assembled for our study, data from the beginning of operation to 1990 are archived for each station. The number of stations has varied in time, a significant increase having occurred at the end of the 1940s. It is only at the beginning of the 1950s that an average density of about one station for 10 000 km2 was reached. We therefore chose to start our regional study in 1951. In the entire database, the total number of stations with less than 10% of missing data is 469 during the first period covered by our study (wet period P1, 1951–70). During the following 20 yr (dry period P2, 1971–90) this number decreased to 433. This decrease is essentially linked to the difficulty of getting data after 1984. This is especially the case for stations not present in the AGRHYMET database. The distribution of stations is the most homogeneous over the 16° × 12° window shown in Fig. 3. It is also a window with a relatively stable number of stations: 323 stations with less than 10% of missing data during the period P1, and 326 stations with less than 10% of missing data during the period P2. However this equality of numbers is somewhat misleading since, as seen in Fig. 3, there are several P1 stations that do not belong to the P2 dataset. This is compensated by an approximately equal number of P2 stations not belonging to the P1 dataset. Figure 3 also shows that the territories of Liberia, Guinea, Ghana, and northern Mali are only sparsely covered (in fact no data were collected for Liberia). Therefore, in the following, the analysis will focus on a window restricted to 4.5°–14.5°N in latitude and 10°W–4°E in longitude, in order to eliminate border effects, especially to the north and west of the 16° × 12° window. This restricted window still samples the full range of West African climates (from south to north: Guinean, Sudanian, and Sahelian). It contains 277 stations with less than 10% of missing data during the period P1 and 299 during the period P2. This latter number corresponds to an average density of one station for 5600 km2. In this 14° × 10° window, the distribution of stations is more regular than anywhere else in the region. However, there remains a central region of low density corresponding to Ghana.

Fig. 3.

Daily rain gauge network. Isohyets refer to the mean interannual rainfall (in mm) for the period 1951–90

Fig. 3.

Daily rain gauge network. Isohyets refer to the mean interannual rainfall (in mm) for the period 1951–90

3. From daily rainfall data to rain event statistics

Rainfall is by nature an intermittent process. In regions where convection is the dominant mechanism producing rain, as is the case in the Tropics, the rainfall regime is a succession of storms lasting a few hours (see D'Amato and Lebel 1998, for statistics of storm duration in the Sahel) and separated by longer interstorm periods. The associated rain fields are highly variable in space. The high space–time resolution of the Estimation des Précipitation par SATellite (EPSAT)–Niger dataset allows us to see how this spatial variability is maintained on timescales far longer than the storm scale (see Lebel et al. 1997; Lebel and Le Barbé 1997). Recent data have confirmed these results. For instance, in 1998, two seasonal totals of, respectively, 450 and 1050 mm were observed at two stations located 80 km apart in a similar environment. The hydrological cycle is dominated by this intermittency. It must therefore be taken into account in hydrological models.

Daily rainfall series provide a biased representation of rainfall intermittency in time. This is clear from the statistics of the number of rainy days and rain events (storms) shown in Table 1. Using a high-resolution record of rainfall at Niamey from an EPSAT–Niger gauge, it can be seen that in August the number of rain events is larger than the number of rainy days, but it is the opposite in June. This derives from the probability of having two events in one day being larger than the probability of having a rain event spread over two different days. In June, it is the opposite. More generally the ratio between the number of rain events and the number of rainy days will depend on the location and period of the year considered. In Niamey for instance the diurnal cycle of rainfall is controlled by local convection, and by the life cycle of large mesoscale convective systems (MCSs) that originate several hundred kilometers eastward in an area bounded by the Joss Plateau and the Air Mountains. D'Amato and Lebel (1998) have shown that during the core of the rainy seasons large MCSs are the main source of rain in the region—this was confirmed by a recent study of Mathon and Laurent (2001)—while in the margins of the rainy season local rain is dominant. The statistics of Table 1 reflect this situation.

Table 1.

Comparison of the number of rainy days and of the number of rain events observed at the Niamey–ORSTOM recording rain gauge station. Statistics are averages over 9 EPSAT–Niger years of measurement (1990–98)

Comparison of the number of rainy days and of the number of rain events observed at the Niamey–ORSTOM recording rain gauge station. Statistics are averages over 9 EPSAT–Niger years of measurement (1990–98)
Comparison of the number of rainy days and of the number of rain events observed at the Niamey–ORSTOM recording rain gauge station. Statistics are averages over 9 EPSAT–Niger years of measurement (1990–98)

In order to obtain a more accurate characterization of the time distribution of rain than that available directly from daily rain series, Le Barbé and Lebel (1997a)—hereafter LBL97—have proposed the use of a Compound Poisson Process (CPP) model allowing for the disaggregation of a daily rain series into a rain event series (see, e.g., Rodriguez-Iturbe et al. 1984, for a discussion on the modeling of temporal rainfall, and Bo et al. 1994, for an earlier work on the disaggregation of rainfall statistics from daily to event scales). In this model, rainfall is represented as a succession of rain events. The duration separating the starting times of two consecutive events is exponentially distributed. The rainfall accumulation H over one event also follows an exponential distribution. The distribution function of accumulations over a period T, denoted HT, is the following:

 
formula

where I1(2nTHT) is the modified first-order Bessel function.

The distribution is characterized by two parameters, nT and h. While h is a scale parameter invariant with the accumulation period T considered, nT is a recurrence parameter, denoting the average number of events over period T. As such, nT is additive, that is for a period T′, it becomes

 
nTTTnT
(2)

Given its additive property, any rainfall series obeying a CPP with white noise exponential distribution may be characterized by its average number of rain events day−1—denoted n in the following—and by h, which is the average rainfall per event, expressed in millimeters. These two parameters fully describe the rainfall regime at a given point in space. The distribution frequency of the rain depths, HT, for any period of accumulation, T, can be computed from these two parameters. Obviously, n and h vary during the course of the year, that is, they are functions of time t and should be denoted n(t) and h(t). In the Sahel, for instance, the peak of the rainy season occurs in August and corresponds to a greater occurrence rate of the rain events. One desirable property of the CPP models is that it holds for nonstationary processes. This is because the sum of two Poisson processes with two different occurrence rates is still a Poisson process. Based on this property and in order to account for the time nonstationarity of rainfall in the region, LBL97 have proposed a moving window method for inferring the parameters n(t) and h(t).

LBL97 successfully used and validated the CPP model on 35 rainfall series in central Sahel. An example of their results is given in Fig. 4 for the station of Niamey, Niger. The model was run with a moving window of 10 days, leading to computed time-averaged values of n and h, plotted in Fig. 4 as a function of t (middle and bottom panels). This is a more relevant representation of the intraseasonal evolution of the rainfall regime, than the sole 10-day hyetogram1 also shown in Fig. 4 (top panel). It is clearly seen by comparing these three graphs that it is the intraseasonal and interdecadal fluctuations of n that control the seasonal cycle of rainfall and its deficit during the dry period. In contrast, the event rainfall seasonal cycle is almost invariant between the two 20-yr periods considered.

Fig. 4.

The rainfall regime of Niamey (periods P1 and P2 treated separately): (top) Rainfall accumulations over successive periods of 10 days; (middle and bottom) The CPP process parameters—mean number of event per day and event rainfall—averaged over the same periods of 10 days, respectively

Fig. 4.

The rainfall regime of Niamey (periods P1 and P2 treated separately): (top) Rainfall accumulations over successive periods of 10 days; (middle and bottom) The CPP process parameters—mean number of event per day and event rainfall—averaged over the same periods of 10 days, respectively

Building on LBL97 (see also Le Barbé and Lebel 1989, for a preliminary test on rainfall series in a more humid tropical climate), it was decided to apply the CPP model to all the series of our 14° by 10° window. The fitting method originally proposed in LBL97 has been refined in order to obtain stable and smoothed parameters. Also, the size of the moving window used to infer the parameters of the model was increased to 15 days. This window size represents an optimum when working on a 20-yr-long series, which is the length of the series used below. A shorter window makes the estimates somewhat unstable. A longer window smoothes out fluctuations of the parameters too strongly with time. The model parameters n(t) and h(t) are thus represented by a discrete series of 24 parameters n15(i) and h15(i), where n15 and h15 denote averages of n(t) and h(t) over 15 days and i is a time index (i = 1, 24). In the figures shown in section 4, the values n15(i) and h15(i) are centered on day number 8 + (i − 1) × 15. In the following, unless other specified, the subscript 15 will be omitted since we will be working only on 15-day averages of n(t) and h(t). A complete presentation of the algorithm may be found in Tapsoba (1997).

The choice—first introduced by LBL97—of the two periods P1 and P2 selected as references for wet conditions followed by dry conditions was followed here because it has the great advantage of producing two periods of equal length of time, which makes statistical comparisons simpler. Of course, the date of separation between the wet period and the dry period is somewhat arbitrary, but this does not seem to constitute a serious issue since the passage from wet to dry years, even if not totally synchronic over the study area, was fairly rapid.

The procedure described above for parameter estimation was identically applied to each of the 277 series of period P1 and to each of the 299 series of period P2. Then for each parameter and each of the 24 time steps, a spatial interpolation using an “error kriging” technique (see Gurascio et al. 1975) was carried out in order to create regular grids. The grid node spacing is 0.25° by 0.25°. Each resulting set of 24 grids may be seen as a space–time cube of resolution 0.25° by 0.25° by 15 days. Four such cubes are available (two parameters, n and h, times two periods, P1 and P2). The combination of the two cubes of period Pi describes the rainfall regime for that period. In Fig. 5, 2D cross sections of the P1 cubes are compared with maps of original data statistics, that is the daily rainfall accumulated over one month, and the number of rainy days for that month.

Fig. 5.

Spatial representation over the study area of the rainfall parameter for the month of Aug (period P1, 1951–70). (top left) Rainfall, (top right) number of rainy days, (bottom left) number of events, and (bottom right) mean rainfall per event. Note the differences between the map of the number of rainy days and the map of the number of rain events

Fig. 5.

Spatial representation over the study area of the rainfall parameter for the month of Aug (period P1, 1951–70). (top left) Rainfall, (top right) number of rainy days, (bottom left) number of events, and (bottom right) mean rainfall per event. Note the differences between the map of the number of rainy days and the map of the number of rain events

Comparing the map of the number of rainy days with the map of the number of rain events n31 (i.e., cumulated over the 31 days of August) illustrates how the model permits us to obtain a better representation of the time intermittency of the rain process. Over the Sahel, the number of rain events is larger—approximately by a factor of 125%—than the number of rainy days. Moving southward, the number of rain events become equal to the number of rainy days around 9°N. Farther south, the number of rainy days becomes far larger than the number of events. This is well in agreement with the known climatology of the convective rain events of the region. As noted previously, in the Sahel in August, most of the rain events are associated with convective systems with a separation time that can be less than one day. On the coast, at that time of year there are relatively few rain events but they are of a different nature and the probability of having a rain event spread over two consecutive days is relatively large.

4. Dynamics of the rainy season

Figure 5 is a 2D-space cross section in the rainfall regime cubes, with a time integration of 1 month. One can also build time–space cross sections for selected longitudes. An example of such cross sections, computed here for the period P2 (the corresponding P1 cross sections will be shown in section 5 below), is given in Fig. 6. On the left the longitude was set to 5°W. On the right it is equal to 2°E. The values so obtained are averages over one grid mesh in longitude, that is 0.25°. These cross sections will be used as a basis for analysing the dynamics of the rainy season in West Africa by looking at the following: i) the onset of the rainy season; ii) the identification of weather zones; iii) the role of the topography.

Fig. 6.

Space–time diagrams of the rainfall parameters represented as cross sections at 5°W (left) and 2°E (right); period P2. Note the strong correlation between n (number of events day−1) and H (rainfall in tenths of mm day−1) at both long. The numbers on the x axis indicate the start of the month (i.e., the representation starts 1 Feb and ends 31 Oct)

Fig. 6.

Space–time diagrams of the rainfall parameters represented as cross sections at 5°W (left) and 2°E (right); period P2. Note the strong correlation between n (number of events day−1) and H (rainfall in tenths of mm day−1) at both long. The numbers on the x axis indicate the start of the month (i.e., the representation starts 1 Feb and ends 31 Oct)

a. The onset

The rain onset may be analyzed by looking either at the n maps or the H maps in Fig. 6, H being the interannual average of the 15-day rainfall interpolated at the nodes of the grid (H is expressed in mm day−1). Note that, due to the algorithm used to compute n and h, we have

 
Hnh.
(3)

The n map at both longitudes presents a similar pattern for the start of the rainy season. First, there is a progressive increase of the number of rain events, propagating northward at an average rate of 2.5° month−1. Note however that, near the coast, the isolines are almost vertical, and the reinforcement of the rainy season is synchronic up to 8°–9°N. That is, the peak in rain event frequency occurs during the first half of May and the decrease occurs simultaneously between 8°N and the coast. During this phase of the onset, the patterns of the n maps and the H maps are similar (cf., e.g., the isoline 0.3 of the n map with the isoline 40 of the H map at 5°W). A striking feature in both series of maps is that the onset of the Sahelian rainy season is not in continuity with the onset of the first rainy season in the south. This is apparent on the n maps and H maps, but especially clear on the latter, when looking at the 5 mm day−1 isoline (numbers in the figure are in 1/10 mm day−1). There seems to be in fact two different dynamics. One is the onset of the rainy season that starts on the coast in February and is propagating regularly northward, reaching 13°N in May. The other is an abrupt surge, in mid-June, of intensified rain occurring synchronously between 9° and 13°N. This space–time cell of high rain is clearly not connected with the space–time cell of high rain on the coast described above.

One main conclusion seems to arise from the above analysis. The onset dynamics does not follow the simple scheme of a regular northward movement with a time lag following the sun zenith position. On the other hand, the movement of the zone of high rainfall toward the coast in September and October seems to follow the sun's movement. This implies a need to reconsider the classic scheme of an undisturbed pattern of weather zones moving north and south with the sun.

b. The annual cycle

According to the pioneering work of Hamilton and Archbold (1945), the West African climate is characterized at any given time by four different weather zones denoted A, B, C, D from north to south and stretching from east to west. The zones move with the sun, subject to a time lag. There are two main features resulting from this description: i) the regular northward migration of the weather zones following the sun produces a progressive onset of rain; ii) the seasonal cycle over West Africa is a progressive shift from a two rainy season regime on the coast to a single rainy season regime in the north. The little dry season may last for a few weeks on the coast. Its length decreases rapidly to the north and it is not felt much beyond 8°N in our study area. Accordingly, the area of rain maximum starts from the coast, moves to the north, reaches its northern limit in the Sahel and then retreats to the south. In a Hovmöller diagram such as the one in Fig. 6, the two points above imply that

  1. on average, the starting date of the rainy season should regularly increase when moving northward;

  2. the nucleus of rain maximum should be continuous and appear as an inverted V shape.

This is clearly not the case, as may be seen from either the n maps or the H maps of Fig. 6. The nonlinear dynamics of the rain zones is very apparent when looking at the 0.3 and 0.4 isolines in the n maps or at the 40 (4 mm day−1) and 50 (5 mm day−1) isolines in the H maps. Strikingly, the first nucleus of rain maximum on the coast is disconnected from the Sahelian rain peak. On the other hand, in agreement with the classic vision summarized above, the Sahelian rain peak is directly connected to the second rainfall peak to the south. The clear cut between the first rainfall peak on the coast and the Sahelian rainfall peak implies that there is no gradual transition from a two rainy season regime in the south to a single rainy season regime in the north. Rather, as illustrated in Fig. 7, one can identify three rain maxima for the regions located between 7°–8°N and 10°N.

Fig. 7.

Daily mean number of rainfall events at three stations lying on a south to north axis around long 2°E, period P2, 1971–90

Fig. 7.

Daily mean number of rainfall events at three stations lying on a south to north axis around long 2°E, period P2, 1971–90

Figure 7 represents the time evolution of n for the stations of Cotonou (5.5°N), Parakou (9.5°N), Benin, and Niamey (13.5°N), all located on the 2°E transect. While Cotonou displays a classic two rain season signal (with, however a secondary “shoulder” before the first maximum), Parakou is characterized by three rainfall maxima occurring around 15 June, 20 July, and 15 September. One interesting result appears when comparing the Cotonou and Parakou signals: the first rain peak happens at the same time at both stations, even though one is located 4°N of the other. A similar feature was found for Abidjan and Yamassoukro in Ivory Coast (not shown). On the other hand, the Niamey maximum, the third Parakou maximum, and the second Cotonou maximum are regularly shifted in time, in agreement with the classic vision of a progressive retreat of the rain maximum to the south, accompanying the ITCZ migration. Another interesting result appears in Fig. 8, where the n signal is plotted for the same three stations, but for the wet period P1. Comparing the curves of period P1 to those of period P2 shows that during the wet period,

  1. the date of the first rain peak in Cotonou and Parakou is unchanged;

  2. the third Parakou maximum and the second Cotonou maximum are delayed by about 10 days in Parakou and 20 days in Cotonou;

  3. in Niamey, the date of the maximum is unchanged, but a secondary maximum appears 20 days before the main maximum; this secondary maximum is synchronous with the second Parakou maximum.

Fig. 8.

Same as Fig. 7, except for period P1, 1951–70

Fig. 8.

Same as Fig. 7, except for period P1, 1951–70

The comparison between Figs. 7 and 8 provides some food for thought regarding how the rainfall regimes were modified during the dry spell, and will be discussed in section 5. Further examination of Fig. 8 suggests a scheme of dry/rainy seasons that departs from the classic vision in the following ways. The first rainy season is the rainy season directly associated with the northward migration of the sun between February and June, which creates a first maximum of received solar energy. The second peak corresponds to a reinforcement of rainfall occurring simultaneously between 9° and 13°N. It should be noted that this is the period where mobile MCSs become the prominent form of rain-generating systems in the region (see D'Amato and Lebel 1998; Mathon and Laurent 2001 for a climatology of convective systems over the Sahelian area). Thus, it appears that, rather than progressively “invading” the northern regions, the zone of deep convection becomes active all of a sudden north of 9°N. Finally, the third peak is associated with the southward migration of the sun and the second maximum in received solar energy.

c. Factors influencing the dynamics of the West African monsoon

The results presented above do not support the classical vision of the West African monsoon (WAM) being driven by a continuous dynamics. While this vision had not been seriously challenged until recently, our finding provides some support to the nonlinear monsoon concept proposed by Eltahir and Gong (1996). This concept is a development, for the case of a moist atmosphere, of the dynamical theory for the response to subtropical thermal forcing of a dry zonally symmetric atmosphere formulated by Plumb and Hou (1992). The scheme of Eltahir and Gong (1996) may explain the existence of the two climatic regimes identified here. The theory states that the meridional distribution of boundary layer entropy controls the partition between a radiative–convective equilibrium regime and an angular momentum conserving regime. While the first corresponds to dry conditions, the second is associated to a healthy monsoon circulation. From our study it may be hypothesized that the brutal jump of the rainfall occurrence rate at the end of June is linked to the achievement of the threshold above which the angular momentum conserving regime becomes effective. When this threshold is reached the meridional circulation develops rapidly and the conditions for rainfall to occur are satisfied over several degrees of latitude. This mechanism is shaping the annual cycle over this region. It is worth noting that the seasonal forecasting method developed by Fontaine et al. (1999) on the basis of the theory proposed by Eltahir and Gong (1996) produces significantly better scores than the alternative stochastic methods that only use various SST predictors. The method is based on the computation of the meridional gradient of moist static energy in April over the region, this gradient being dependent on the SSTs and on the continental surface conditions. There is thus some converging evidence that the surface conditions, both over the ocean and over the continent, strongly modulate the atmosphere dynamics on a regional scale, producing a nonlinear behavior of the WAM.

The relief is another factor that may play a significant role in the dynamics of the WAM. Figure 9 shows the 2D maps of n for 6 dates at 30-day intervals for the dry period P2. The influence of the two mountain ranges, the Fouta-Djalon in the west and the Atakora in the east—crossing Togo and ending in northwest Benin—is clearly visible in the maps of day 165 (mid-June) and day 195 (mid-July). During this first part of the rainy season, the maxima of rainfall occurrences are located on the upwind side (west–southwest) of these two ranges. Associated with these maxima are areas of lower rainfall located on the lee side of the two mountain ranges. Then, after mid-July, the regime of large convective systems moving from east to west becomes prominent and the location of the maxima associated with the ranges is shifted to the east. It seems that in the latter type of circulation regime, the barrier effect of the topography tends to disappear or even to act in the opposite way, shielding the southwest side of the mountains. A very similar pattern is observed for the wet period P1 (not shown). At a larger scale, it is worth noting that a recent modeling simulation experiment (Semazzi and Sun 1997) has pointed to the possible role of the Atlas Mountains and the Ahaggar Plateau in inducing an orographic circulation over the Sahelian region. These mountains are located at the critical latitude of the descending branch of the Hadley cell, the location of which is directly related to the equilibrium conditions in the theory of Plumb and Hou (1992). It consequently cannot be excluded that this unique feature of the West African topography might interact with the equilibrium conditions controlling the monsoon circulation in case of a significant and durable modification of the general circulation, thus resulting in an abrupt climate change over the region. Enhanced and sustained observations over the whole transect extending from the Guinea Gulf to the Sahara are a necessity to control such theories.

Fig. 9.

2D maps of the mean number of rain events day−1 (n) over the study area for the dry period P2. Day 165 is mid-Jun, day 195 is mid-Jul, day 225 is mid-Aug

Fig. 9.

2D maps of the mean number of rain events day−1 (n) over the study area for the dry period P2. Day 165 is mid-Jun, day 195 is mid-Jul, day 225 is mid-Aug

5. Wet spell and dry spell in West Africa

The severity of the drought that struck the Sahel in the years 1971–74 generated several studies looking for factors explaining this climatic event. From the early work of Charney et al. (1977) using GCMs to recent studies using regional models (see, e.g., Zheng and Eltahir 1998) a series of modeling exercices were carried out regarding the potential impact of vegetation degradation on the West African rainfall. The role of the oceans was also investigated (e.g., Folland et al. 1986; Lamb and Peppler 1992; Semazzi et al. 1996; Ward 1998). Despite the improved capabilities of these models, a proper evaluation of the respective roles of the continental surface conditions and of the tropical oceans (and of their interactions as well) remains to be achieved. This will constitute a major challenge for the coming years, as underlined by Nicholson (2000) and in the recent report produced by the CLIVAR Africa Task Team (2000, hereafter CATT). A prerequisite to such studies is to characterize as precisely as possible what changes occur in the rainfall regimes when shifting from the wet to the dry period. The comparative analysis carried out below on the rain event statistics of these two periods is a contribution to that task.

To this end, we compare the maps of Fig. 6—that were analyzed previously—with similar maps (shown in Fig. 10) computed for the wet period P1. For a given longitude, the respective patterns of the n maps of each period are strikingly similar. This is also true for the h maps. Compared to the wet period, the n values of the dry period are decreased by 0.1–0.2 events per day, while the H values are decreased by 1–1.5 mm day−1, but the location and date of the maxima and minima remain unchanged. This is true for both longitudes examined here (5°W and 2°E). On the other hand, the h maps display in all cases some patchiness. It is difficult to identify well-defined structures. While the decrease of n during the dry period is systematic, the areas where h values decreased are globally balanced by areas of increased h. Previous studies focusing on the Sahel (e.g., LBL97; D'Amato and Lebel 1998) have shown that the rainfall variability and the variability of the number of events are strongly correlated at both interannual and decadal timescales. The above comparison confirms these first results and extends them to the Soudanian area of West Africa. At the same time, the n map of period P2 is not a simple homothetic replication of the n map of period P1. In order to visualize the differences between the two periods, two maps (shown in Fig. 11) were constructed. In each map (one for each longitude), the isolines refer to period P2 while the color image is drawn from the P1 values. This representation confirms a significant shift of the position of the second zone of rain maximum, whereas the position of the first zone of rain maximum is only slightly shifted. On the coast the timing of the first rainy season is advanced by about one week during the dry years. As for the timing of the second rain zone, it is seen that its maximum occurs about 20 days earlier at all latitudes between 6° and 10°N. This time shift is progressively decreases to the north: at 14°N it is reduced to about one week. It is remarkable that these values are very similar whether computed on the 2°E map or on the 5°W map. Looking back at the diagrams of Figs. 7 and 8, it is seen that a similar overall trend is observed on point rainfall series. However, the values are slightly different, as mentioned above in section 4. They indicate a regular decrease of the time shift when moving away from the coast (20 days in Cotonou, 10 days in Parakou, and no lag in Niamey). Given the smoothing effect produced by the mapping of Fig. 11, it is not possible at this point to ascertain whether the early timing of the second rainy season during the dry years must be considered as constant over the 6°–10°N region or whether, as indicated by Figs. 7 and 8, there is a zonal gradient.

Fig. 10.

Space–time diagrams of the rainfall regime represented as cross sections at 5°W (left) and 2°E (right). Period P1. As in Fig. 6, note the strong correlation between the n (number of events day−1) and the H (accumulated rainfall in mm) at both long. Also note the similarity of the patterns between the two figures. During the dry years, the cumulative rainfall and the number of events decreased but the general space–time pattern was preserved

Fig. 10.

Space–time diagrams of the rainfall regime represented as cross sections at 5°W (left) and 2°E (right). Period P1. As in Fig. 6, note the strong correlation between the n (number of events day−1) and the H (accumulated rainfall in mm) at both long. Also note the similarity of the patterns between the two figures. During the dry years, the cumulative rainfall and the number of events decreased but the general space–time pattern was preserved

Fig. 11.

Space–time distributions of the number of events obtained for period P1 and P2 are superimposed (top) Cross section at 2°E, (bottom) cross section at 5°W). The distribution for period P1 is represented by the color image while the distribution for period P2 is represented by the contour map. This representation highlights the time shift of the rainy seasons. While this shift remains moderate for the first rainy season in the south it amounts to almost 15 days for the peak of the Sahelian rainy season

Fig. 11.

Space–time distributions of the number of events obtained for period P1 and P2 are superimposed (top) Cross section at 2°E, (bottom) cross section at 5°W). The distribution for period P1 is represented by the color image while the distribution for period P2 is represented by the contour map. This representation highlights the time shift of the rainy seasons. While this shift remains moderate for the first rainy season in the south it amounts to almost 15 days for the peak of the Sahelian rainy season

A summary of the main changes that occurred in the rain event statistics between the wet and dry seasons is as follows.

  1. Even though changes in the mean event rainfall (h) happened here and there, these changes are not spatially organized; areas of decreased h are balanced by areas of increased h and, when expressed in relative values, the range of these changes are small.

  2. In contrast, changes in the mean number of event (n) are extremely well organized in space. A decrease is observed systematically. It accounts for most of the global rainfall deficit between the two periods except possibly on the coast for the little dry season.

  3. The deficit of the number of rain events is most pronounced during the core of the rainy season in the Sahel and during the first rainy season on the coast.

  4. The total length of the rainy season is almost unchanged in the Sahel. Farther south, the date of onset is not modified much, but the second rainy season starts earlier and finishes earlier. In total the length of this second rainy season is diminished, but by no more than 10 days.

  5. There is a very distinctive and global time shift of the second rain zone extending from the Sahel to the coast between August and October. The whole structure is shifted by 10–20 days with its maximum occurring earlier during the dry years.

The impact of the drought on the hydrological cycle and the agriculture are not independent of the above changes. In the Sahel the greatest threat to agriculture seems to be an increased risk of dry spells during the core of the rainy season at periods where the millet may be most sensitive to it. On the other hand, a reduction in the length of the rainy season does not appear to be very significant and thus the quest for new grain varieties with a shorter growing cycle may not be so necessary. From a hydrological point of view, the increase in the mean time between two successive rain events, with little changes in the mean event rainfall probably implies that, all other things being equal, the average resulting runoff might not change much. Since the soil dries out rapidly in the Sahel, the initial wetness conditions when the rain starts are not critically modified when increasing the average time between two events by 20% or so, or by an increased probability of longer dry spells. The vegetation, however, will suffer. It may thus be inferred that a long series of dry years combined with demographic pressure and the demand for cooking wood, will have a serious impact on the vegetation, thus modifying the soil surface conditions. This, in turn, may change the ratio between rainfall and runoff. There are thus interacting consequences of long lasting droughts that require further attention by both observation programs and climate modelers.

In the south it seems that the deeper impact for the agriculture is the advance in the timing of the second rainy season and its overall shortening. Also, the importance of the first rainy season on the coast is significantly reduced. The combination of these two factors might have significant consequences for agricultural planning.

6. Discussion and perspectives

The main result reported in this study is that analyzing the rainfall regimes of West Africa in terms of rain event occurrence rate and intensity provides an improved characterization of the weather dynamics of this region. In particular, it questions the classic vision of a progressive transition from a two rainy season regime on the coast to a single rainy season in the Sahel. The sudden surge of a maximum of rain event occurrence frequency between 10° and 14°N, starting in mid-June, has, to our knowledge, never been pointed out before. It raises some intriguing questions regarding the equilibrium between atmospheric moisture advection on the continent and energetic factors that may create the required conditions for the formation of fast-moving convective complexes that account for most of the rainfall in July and August at these latitudes. This is in line with the nonlinear monsoon theory developed by Eltahir and Gong (1996) and the recent work of Fontaine et al. (1999) that shows that seasonal forecasts of the rainy season over West Africa may be improved by including the horizontal moist static energy gradient as a predictor.

Another important result, already found for the Sahel by Le Barbé and Lebel (1997a) is that most of the rainfall deficit of the dry period 1971–90 is correlated with a general decrease of the occurrence rate of rain events. This, along with the fact that the rainfall deficit was equal to 180–200 mm all over West Africa, indicates that there was no crucial difference in nature between the Sahelian drought and the drought of the Soudano–Guinean regions. It does not appear that the explanation of such features resides in an abnormally southward position of the ITCZ during the dry spell.

Given the method used in this paper it was not possible to investigate the interannual variability of the West African rainfall. While D'Amato and Lebel (1998) have shown from high-resolution rain data that in the Sahel, the interannual variability of the seasonal rainfall was also related to the variability in the number of mesoscale convective systems, other works have pointed to other factors. Janicot (1992) has shown that distinguishing two regions north and south of 10°N, allows a better description of the interannual variability of the West African rainfall. More recently Janicot et al. (1996) have shown that the correlation field between the world ocean SSTs and the Sahelian rainfall, computed at the interannual scale, changed markedly at the turning point of the end of the 1960s. Another important point to consider from an hydrological perspective is the occurrence of extreme rainfall. The global model and approach used here did not permit us to tackle that question. It is not clear whether fluctuations in the magnitude and occurrence of extreme events are similar, whether looking at the interannual variability or focusing on the contrast between the two periods analyzed here. There obviously remain room for work in this respect. It should be noticed, however that the data currently available in West Africa are not sufficient to analyze in detail the various modes of variability of rainfall. It is thus important to obtain accurate and high-resolution data that will allow us to study the intensity, spatial extension, and structure of the rain fields at various timescales in order to better characterize their variability from the intraseasonal to the decadal scales. Specific observing systems will be required for that purpose, as recommended by CATT (2000). Such systems will also be extremely useful in validating satellite estimates and model outputs.

Presently, simulations by global or regional models are not able to reproduce the rainfall variability at the scale of the rain event. GCM resolution is too coarse for that purpose and their internal variability is far too large (see, e.g., Semazzi et al. 1996). It is thus an important goal for modeling studies to be able to identify rain events in regional models realistically, so as to compare their statistics to those of the observations. It is necessary to attain the rain event scale in climate modeling in order to study the impact of climate variability on the water resources under various climatic scenarios.

Acknowledgments

We are grateful to the various national meteorological services that helped the Institut de Recherche pour le Développement (IRD, formerly ORSTOM) to build the daily rainfall database used in this study. Special thanks are due to Elfatih Eltahir and an anonymous reviewer for constructive comments on section 4 of the paper, and to Nick Hall for helping to put it in proper English. This research was funded by IRD.

REFERENCES

REFERENCES
Bo
,
Z.
,
S.
Islam
, and
E. A. B.
Eltahir
,
1994
:
Aggregation-disaggregation properties of a stochastic rainfall model.
Water Resour. Res
,
30
,
3423
3435
.
Charney
,
J. G.
,
W. J.
Quirk
,
S. H.
Chow
, and
J.
Kornfield
,
1977
:
A comparative study of the effects of albedo change on drought in semi-arid regions.
J. Atmos. Sci
,
34
,
1366
1385
.
CLIVAR Africa Task Team
,
2000
:
CLIVAR-Africa Implementation Plan.
CLIVAR, C. Thorncroft, Ed., 33 pp. [Available online at http://www.met.rdg.ac.uk/-swsthcri/catt.html.]
.
D'Amato
,
N.
, and
T.
Lebel
,
1998
:
On the characteristics of the rainfall events in the sahel with a view to the analysis of climatic variability.
Int. J. Climatol
,
18
,
955
974
.
Eltahir
,
E. A. B.
, and
C.
Gong
,
1996
:
Dynamics of wet and dry years in West Africa.
J. Climate
,
9
,
1030
1042
.
Folland
,
C. K.
,
T. N.
Palmer
, and
D. E.
Parker
,
1986
:
Sahel rainfall and worldwide sea temperature 1901–1985.
Nature
,
320
,
602
607
.
Fontaine
,
B.
, and
S.
Janicot
,
1996
:
Sea surface temperature fields associated with West African rainfall anomaly types.
J. Climate
,
9
,
2935
2940
.
Fontaine
,
B.
,
S.
Trazaska
, and
S.
Janicot
,
1998
:
Evolution of the relationship between near global and Atlantic SST modes and the rainy season in West Africa: Statistical analyses and sensitivity experiments.
Climate Dyn
,
14
,
353
368
.
Fontaine
,
B.
,
N.
Philippon
, and
P.
Camberlin
,
1999
:
An improvement of June–September rainfall forecasting in the Sahel based upon region April–May moist static energy content (1968–1997).
Geophys. Res. Lett
,
26
,
2041
2044
.
Gurascio
,
M.
,
M.
David
, and
C.
Huijbregts
,
1975
:
Advanced Geostatistics in the Mining Industry.
NATO Advanced Study Institutes Series, Vol. C24, Reidel Publishing, 461 pp
.
Hamilton
,
R. A.
, and
J. W.
Archbold
,
1945
:
Meteorology of Nigeria and adjacent territory.
Quart. J. Roy. Meteor. Soc
,
71
,
231
265
.
Hastenrath
,
S.
,
1990
:
Decadal scale changes of the circulation in the tropical Atlantic sector associated with Sahel drought.
Int. J. Climatol
,
20
,
459
472
.
Janicot
,
S.
,
1992
:
Spatiotemporal variability of West African rainfall. Part I: Regionalizations and typings.
J. Climate
,
5
,
489
511
.
Janicot
,
S.
,
V.
Moron
, and
B.
Fontaine
,
1996
:
Sahel droughts and ENSO dynamics.
Geophys. Res. Lett
,
23
,
515
518
.
Lamb
,
P. J.
,
1983
:
West African water variations between recent contrasting Subsaharan droughts.
Tellus
,
35A
,
198
212
.
Lamb
,
P. J.
, and
R. A.
Peppler
,
1992
:
Further case studies of tropical Atlantic surface atmospheric and oceanic patterns associated with sub-Saharan drought.
J. Climate
,
5
,
476
488
.
Le Barbé
,
L.
, and
T.
Lebel
,
1989
:
Analysis of regional precipitation patterns in West Africa based on the law of leaks.
Proc. Fourth Int. Meeting on Statistical Climatology, Rotorua, New Zealand, Amer. Meteor. Soc. and New Zealand Statistical Association, 273–277
.
Le Barbé
,
L.
, and
T.
Lebel
,
1997a
:
Rainfall climatology of the HAPEX-Sahel region during the years 1950–1990.
J. Hydrol
,
188–189
,
43
73
.
Le Barbé
,
L.
, and
T.
Lebel
,
1997b
:
Variabilité interannuelle et décennale de la pluviométrie en Afrique de l'Ouest: Diagnostic et modélisation.
(Interannual to decadal variability of rainfall in West Africa: Diagnosis and modeling). Preprints, Climate Variability Prediction, Water Resources and Agricultural Productivity: Food Security Issues in Sub-Saharan Africa, Cotonou, Benin, ORSTOM, SCOWAR, START, and WCRP
.
Lebel
,
T.
, and
L.
Le Barbé
,
1997
:
Rainfall monitoring during HAPEX-Sahel: 2. Point and areal estimation at the event and seasonal scales.
J. Hydrol
,
188–189
,
97
122
.
Lebel
,
T.
,
J. D.
Taupin
, and
N.
D'Amato
,
1997
:
Rainfall monitoring during HAPEX-Sahel: 1. General rainfall conditions and climatology.
J. Hydrol
,
188–189
,
74
96
.
Mathon
,
V.
, and
H.
Laurent
,
2001
:
Life cycle of the Sahelian mesoscale convective cloud systems.
Quart. J. Roy. Meteor. Soc
,
127
,
377
406
.
Nicholson
,
S. E.
,
1981
:
Rainfall and atmospheric circulation during drought periods and wetter years in West Africa.
Mon. Wea. Rev
,
109
,
2191
2208
.
Nicholson
,
S. E.
,
2000
:
Land surface processes and Sahel climate.
Rev. Geophys
,
38
,
117
139
.
Plumb
,
R. A.
, and
A. Y.
Hou
,
1992
:
The response of a zonally symmetric atmosphere to subtropical thermal forcing: Threshold behavior.
J. Atmos. Sci
,
49
,
1790
1799
.
Polcher
,
J.
,
1995
:
Sensitivity of tropical convection to land surface processes.
J. Atmos. Sci
,
52
,
3143
3161
.
Rodriguez-Iturbe
,
I.
,
V. K.
Gupta
, and
E.
Waymire
,
1984
:
Scale considerations in the modeling of temporal rainfall.
Water Resour. Res
,
20
,
1611
1619
.
Semazzi
,
F. H.
, and
L.
Sun
,
1997
:
The role of orography in determining the Sahelian climate.
Int. J. Climatol
,
17
,
581
596
.
Semazzi
,
F. H.
,
B.
Burns
,
N. H.
Lin
, and
J. K.
Schemm
,
1996
:
A GCM study of teleconnections between the continental climate of Africa and global sea surface temperatures anomalies.
J. Climate
,
9
,
2480
2497
.
Tapsoba
,
D.
,
1997
:
Caractérisation événementielle des régimes pluviométriques Quest-Africains et de leur récent changement.
Thése de l'Université Paris XI, 145 pp
.
Taylor
,
C.
, and
T.
Lebel
,
1998
:
Observational evidence of persistent convective scale rainfall patterns.
Mon. Wea. Rev
,
126
,
1597
1607
.
Ward
,
M. N.
,
1998
:
Diagnosis and short-lead time prediction of summer rainfall in tropical North Africa at interannual and multidecadal timescales.
J. Climate
,
11
,
3167
3191
.
Zheng
,
X.
, and
E. A. B.
Eltahir
,
1998
:
The role of vegetation in the dynamics of West African monsoons.
J. Climate
,
11
,
2078
2096
.

Footnotes

* Current affiliation: Direction de l'Hydraulique du Benin, Cotonou, Benin.

+ Current affiliation: Chaire industrielle en hydrologie statistique, INRS-Eau/Hydro-Québec/CRSNG, Sainte-Foy, Quebec, Canada.

Corresponding author address: Dr. Thierry Lebel, LTHE, BP 53 F-38041, Grenoble Cedex 9, France. Email: lebel@hmg.inpg.fr

1

Rainfall accumulation over a successive periods of days.