Agroclimatological Characteristics of Rainy Seasons in Southwestern Burkina Faso during the 1970-2013 Period

The different drought phases observed during the 1970-2010 period have underlined important weaknesses of West African agricultural systems. The droughts resulted in important decreases in crop production, triggering a significant deficit in food availability. Many studies have identified changes in rain events seasonal patterns as the key drivers of agricultural production failure during these drought phases. In this study, seven agriculturally-relevant intra-seasonal rainfall characteristics (i.e., annual rainfall amount, onset and cessation of the rainy season, dry spells, extreme rain events, hot spells, and strong winds) and associated constraints to crop growth are described for the main cereals (maize, millet, and sorghum) in southwestern Burkina Faso. These characteristics are calculated or determined using daily climate data from a local network of 16 weather stations spanning the 1970-2013 period. A computation of the intensity and the occurrence of these phenomena during the rainy seasons helped to draw the rainy seasons’ nomenclature. Findings suggest that the rainy seasons during the drought phases are characterized by low annual rainfall amount, late onset, early cessation and more frequent long dry spells (>7 days). Furthermore, the long dry spells mostly supplementary irrigation systems during long dry spells.


Introduction
The extreme drought events recorded across West Africa in the 1970s/80s have revealed important weaknesses of agricultural systems in the region [1] [2] [3] [4]. A number of several rainy seasons with important crop failure, resulting in significant decreases in food production were recorded [5] [6] [7]. As proposed by many studies [8] [9] [10] [11], this decrease in food production is mainly attributed to the deterioration of climatic conditions, i.e. significant changes in the regional rainfall regime [12]- [17]. More specifically, investigations pointed out climatic factors such as delayed rainy season onsets, a decrease in the annual rainfall amount, a decrease in the number of rain events, and long dry spells.
Previous studies concluded that the timing of rain events during the rainy season significantly determines crop yield [17] [18] [19]. Comparing two rainy seasons (1991 and 1992) similar with respect to annual rainfall amount, [20] found from the observations within a dense rain gauge network in Niger, different timings of rain events resulting in different impacts on agricultural production. For instance, in 1992, the rainy season produced more dry spells of more than five days long, which led to reduced millet crop yields and very low development of grass cover. It is therefore critical to conduct an analysis of rainy season's governing characteristics of crop growth to help better describe the potentialities and constraints during the growing season. From the existing literature, dry spells are often listed as the most crucial characteristic of the West African rainy season that influences crop yield [17]. Depending on its duration and timing during the growing season, a dry spell of more than ten days can reduce maize yield by more than 70% [21]. Such a magnitude of crop production decrease is likely to result in a regional food shortage. Similarly, Waonga et al. [19] showed from an assessment of the impacts of the planting date on maize yield, that the planting date has a great impact on maize production during the rainy seasons in Burkina Faso. They found that with an optimized planting date (OPD), the potential maize yield would increase by 15%. Furthermore, other characteristics such as hot spells and strong winds could hamper crops development or damage crops at any development stage [11] [22] [23]. Importantly, these factors are being exacerbated by the observed global warming in the region, manifested through a significant increase in temperatures [24] [25]. In the context of global warming combined with increasing population pressure and frequent yield losses due to erratic rainfall in West Africa, there is a need to improve the stability and productivity of agricultural systems. In this study, using only in-situ observation datasets (daily data), to avoid systematic biases that occur in some of the state-of-the-art climate products [26], data of three main climate variables (rainfall, temperature and wind velocity) are collected and analyzed to derive major agriculturally relevant rainfall characteristics in West Africa: annual rainfall amount, onset and cessation of the rainy season, dry spells length, heavy rain events, hot spells length, and strong wind velocity. The probability of occurrence, the detrimental threshold, the timing of the different climatological characteristics and their inter connections in south- and Adapted Land use) intervention zones in West Africa, is structured as follows: In Section 2, we describe the study area and datasets. In Section 3, we provide an overview of the different methods adopted. In Section 4, results are presented and conclusions are drawn in Section 5.

Study Area and Dataset
The study is carried out in southwestern Burkina Faso, around Dano (Figure 1 (i.e. a significant trend in a time series is testified by a p-value lower than 5%).

Description of the Rainy Season Characteristics
Over the region, the single rainy season from April to October is characterized by two phases: the first phase (establishment) called onset of the rainy season and the second phase (retreat) referred to as the cessation of the rainy season.
Several methods were developed to determine onset and offset dates of rainy seasons in West Africa from daily rainfall amounts [9] [16] [34]. In this study, four different methods are used to determine the two main dates for each year.
First, the agronomic method proposed by Balme et al. [35] from the original method developed by different research groups (e.g. [9] [16] [34]), identifies the two dates based on the following rules:  The onset is defined by three days of a cumulative rainfall amount higher than 20 mm after April 1st, with no dry spell of more than seven days within the following 30 days (three dekads);  The offset is defined by the last rainfall higher than 5 mm/day after September 1st and with no rainfall higher than 5 mm/day during the following 20 days (two dekads).
The three other methods used to identify rainy seasons onsets are defined by the amount of the daily rainfall events as proposed by [35] for the hydrology approach. The onset is defined by the first rainfall event in the year with an amount higher than a threshold without considering dry spells after the rainfall event.
Three different threshold values were considered: 5 mm/day, 10 mm/d and 20 mm/d. The rainy season's offset remains the same as previously described because this characteristic is very stable from one method to another according to Ibrahim et al. [34].
The other characteristics of the rainy season are determined during the period defined by these two salient dates. Nonetheless, we consider the annual rainfall amount which includes rainfall events recorded outside the rainy season period instead of the seasonal rainfall. Though the two quantities present the same inter-annual variability and a difference in magnitude lower than 10% [36]. Dry spells during the rainy season are determined with the rainfall threshold of 1 mm/d. Thus, a day is considered dry if the recorded rainfall amount is less than  [21] show that dry spells become harmful to crops when they last more than seven days. Temperature is the second climate parameter that influences crop development. Gourdji et al. [37] presented critical threshold values of high temperatures for crop growth: 34˚C for wheat, 35˚C for maize, 36˚C for rice and 39˚C for soybeans. The hot spells are defined by more than three consecutive days with daily maximum temperature higher than 35˚C.

Statistical Distribution of Extreme Climate Events
Extreme climate events include unusual, severe or unseasonal weather; weather conditions at the extremes of the historical distributions-the range that was not observed in the past [38]. The designation of extreme climate events depends on the region and the objectives of the study [39] [40] [41]. Beniston and Stephenson [40] stated that "there is no single definition of what constitutes extreme events" and Stephenson [41] suggested that the main attributes of extreme events can be defined from the following: rate of occurrence (probability per time unit);  magnitude (intensity);  duration and timing;  spatial scale (footprint);  multivariate dependencies.
Extreme climate events are also characterized by their impact on human wellbeing as they generally entail important economic loss and mortality.
Three climate extreme events are considered in this study: heavy rain events, long dry spells and strong winds. The first extreme event often results in floods (drowned cropland); the second is the source of crop failure and the last extreme event is the source of important damages. The intensity and the occurrence of extreme events are described with the Gumbel distribution after the method by El Adlouni et al. [42]. This method helps to relate events' intensity to their frequency of occurrence and vice versa through the probability distribution. The intensity of extreme events is inversely proportional to the frequency: 1 Intensity frequency of occurence ∝ The probability density function (pdf) of the Gumbel distribution is presented by: Let y be the reduced variable with The cumulative distribution function (cdf), denoted F(X < x), is the probability that the random variable X is less or equal to x (non-exceedance probability ( ) F x ). The 100p percentile xp is the value with cumulative probability of p. The 100p percentile is also called the 100(1 − p) percent exceedance event because it will be exceeded with probability 1 − p. The corresponding return period T is: F cumulative distribution function and T the return period. If T is known, T y can be found from Equation (4), and once T y is known, T x can be computed by: Thus, the intensity of x corresponding to different return periods T (and vice versa) can be computed from Equations (4) and (5).

Results
The seven rainfall characteristics (annual rainfall amount, onset and cessation of the rainy season, dry spells, heavy rain events, hot spells, and strong winds) are computed for each rainy season from 1970 to 2013. The inter-annual variability of each characteristic is analyzed at local and regional levels.         years (1973,1983,1984,1990,1992, and 2005) present significant low annual rainfall amount, and another six years (1971, 1986, 1994, 1999, 2003, and 2010) present significant high annual rainfall amount. Figure 7 presents the propor- number of stations affected, shows that five years (1973, 1983, 1984, 1990, and 2005) are extremely dry (a shift of more than 10% and more than 80% of the network affected), and six years (1971,1986,1994,1999,2003, and 2010) are extremely wet (an increase of more than 10% and more than 80% of the network affected Interesting findings are also derived from the analysis of the spatial distribution of annual rainfall during the driest year (1983) and the wettest year (1999).

Inter-Annual Evolution of the Annual Rainfall Amount
For instance, Figure 8 which presents the spatial distribution of annual rainfall deficit in 1983, highlights that this drought episode did not affect all the locations seamlessly. The highest deficit, more than 65%, is recorded around Leo in the East, while it is less than 25% in the West along the axis from Diebougou to Bondoukuy and around Gao in the Northeast. This spatial disparity also appears for the wettest year of 1999 ( Figure 9): less than 20% along the axis from Bagassi to Boura, and higher than 40% around Gao. The two figures (

Distribution and Evolution of Heavy Rainfall Events
The intensities of the annual maximum rainfall events in the study area vary  Figure 10). The highest intensities for all the stations during the 1970-2013 period are higher than 120 mm/d (Table 2). Thus, the extreme intensity (with regard to flood) of 100 mm/d [44] can be exceeded in the study area. The annual heaviest rainfall events are mainly recorded in the heart of the rainy season: from mid-July to end of August. However, isolated intense rainfall events can occur outside this main period: rainfall amounts of 123 mm/d and 110 mm/d were recorded respectively on the 29/04/1998 at Boromo and on the 31/10/1974 at Gaoua.
Furthermore, Figure 11 & Figure 12 show the high spatial disparity that characterizes extreme rainfall intensity distribution in the region during a rain event.   same spatial contrast appears for the heaviest event recorded on 28/07/2007 at Leo in the east with 187.5 mm/d, whereas the Ouo station, 210 km in the Southwest received no rainfall. Table 2 shows that the sixteen stations did not record the highest intensity in the same year during the 1970-2013 period; a maximum number of two stations recorded the same high intensities in 1986, 1989, 1991, 2007 and 2008. In the same way with the annual rainfall amount, the inter-annual evolution of the intensity of the annual heavy rain events presents no significant trend from 1970 to 2013 (p-values > 50%). There is only a slight increase in the mean intensity during the 1986-1999 period in comparison to the 1970-1985 period (figure not shown).
The cumulative density functions of the extreme events are very similar in the region (Figure 13). The cumulative density curves for the three main stations (Gaoua, Dano and Boromo) are very close for low frequency (F > 0.97) and high frequency (F < 0.4). Figure 13 shows that the median intensity of the annual heaviest rain event in the region is about 80 mm/d. This intensity represents the highest intensity recorded in the study area at least once during two consecutive years. The probability that a rain event will not exceed 100 mm/d during the rainy season varies from 72% at Gaoua to 75% at Boromo (Figure 13). This probability means that this threshold could not be exceeded more than once within a period of five years. The one hundred-year and one thousand-year are about 184 mm/d and 250 mm/d, respectively. The recorded highest rainfall of 187.5 mm/d in the region corresponds to a return period of about 150 years.

Dry Spells Timing and Spatial Extension
Dry spells during the rainy season are mentioned in many studies [17] [21] [45] [46] as a key factor hindering crop growth in the Sudano-Sahelian zone. The impacts of dry spells on crops depend on the period of their occurrence during the growing season, as well as their duration and frequency [21]. The probability that a day (occurrence of dry spell divided by the 44 years of the whole period of observation) is part of a dry spell of more than five days is very large (>60%) at the beginning and at the end of the rainy season ( Figure 14). A dry spell of more than two weeks can occur at any time during the season, the probability of their occurrence is higher than once in every ten years around the 1st of May, mid-July and at the end of September. Table 3, showing the mean number of dry spells during the rainy season, displays eight dry spells of 5 to 7 days, three dry spells of 8 to 13 days, and one dry spell longer than two weeks. Figure 15 presents the duration of seasonal longest dry spells underlines those dry spells can even exceed two weeks. Over the 16 stations, the maximum duration (red dots on Figure 15) at each station is higher than 15 days. This duration is reported to be prejudicial to all crops at any stage [21]. The inter-annual average for all 16 stations is about ten days ( Figure 15), but the longest dry spells exceed two weeks in 27 out of the 44 years of observations (Figure 16). From a socio-economic and climatic survey made in the study area in May 2014 [30], majority of farmers emphasized that the tolerance level to dry spells for maize, Table 3    to the decadal dry spell and the three weeks duration dry spell corresponds to one-hundred-year return period dry spell. From Table 4 it appears that dry spells are much longer in the South than in the North: a difference of 3 to 4 days for the different quantiles between the South (Gaoua or Ouo) and the North (Bondokuy or Ouakara).

Description of Hot Spells Timing during the Rainy Season
Hot spells are also considered as an important climate risk for crop development [47]. In fact, long hot spells during the rainy season contribute to increasing crop evapotranspiration, drying out crops and therefore reducing significantly crop yields. In Figure 17, the representation of seasonal maximum temperatures at Gaoua and Boromo, indicates that maximum temperatures are always higher than 38.5˚C. Thus, the critical temperatures for maize and rice were exceeded in  Also, the difference in the two durations of the two lag times is similar to the difference in the lag times between the onset (from South to North) and the offset (from the North to the South) of the rainy season.

Description of Wind Speed during the Rainy Seasons
The mean wind speed in the region is about 1.5 m/s from the measurements at  May. Youm et al. [48] investigating seasonal variation in wind speed over Senegal from a set of four stations. The frequency of strong winds is higher at Gaoua in comparison to Boromo. Table 5, displaying the classification of wind effects from their speeds, shows that the strong winds (above 3 m/s) recorded in the study area would certainly have negatively affected crops.

Summary and Discussions
The southwestern part of Burkina Faso, like all of West Africa [1] [49], has recorded a high inter-annual variability in the annual rainfall amount during the 1970-2013 period. Even if no significant linear trend appears in the annual rainfall time series from 1970 to 2013, significant dry/wet years have been recorded in the region. These extreme years appear along the whole period and there is no decade with only one type of extreme year (dry or wet). Investigation of these extreme years revealed that 1983 and 1984 are the driest consecutive years with −27% and −18% decrease in the annual rainfall amounts, respectively. In contrast, 1999 recorded the highest annual rainfall amount with an increase of +22%. The comparison of rainy season characteristics during dry and wet years ( Table 6) indicated later onsets and earlier cessations in dry years relatively to wet years, resulting in shorter rainy seasons for dry years: about one month of lag difference between the dry years of 1983 and the wet year of 1999 ( Table 6). The number of rainy days in 1983 is also about half the number of rainy days in 1999. Furthermore, there is a lengthening of dry spells and an increase in their frequency during the dry years. However, there is no significant change in the mean amount of daily rainfall despite an increase in the maximum daily rainfall amount for wet years. The seasonal maximum daily amount of rainfall is also not linked to the wetness or dryness of the season: the correlation coefficient between the annual rainfall amounts and the annual maximum daily rain fall amounts is less than 0.3 at all the stations. These results correspond to the find-  Mean dry spell duration (days) 9 8 8 7 7 8 *Regional represent the average from the 16 stations. by the occurrence of dry spells (more than five consecutive days with rainfall amount lower than 1 mm) and hot spells (more than three consecutive days with maximum temperature higher than 35˚C). Both characteristics present high probability at the starting and the ending of the rainy season. However, the correlation between the two characteristics with regard to their daily occurrence is very weak (less than 0.1 at Gaoua and Boromo).

Conclusions
During the last four decades , the rainy seasons in West Africa were characterized by high inter-annual variability. This study on the description of agriculturally relevant rainfall characteristics in southwestern Burkina Faso, during the 1970-2013 period, reveals five extremely dry years (1973,1983,1984,1990, and 2005) and six extremely wet years (1971, 1986, 1994, 1999, 2003, and 2010). Our findings indicate that during dry years, very late onsets and very early offsets, contribute towards lowering their agricultural potentialities through a significant decrease in the amount of rain events and a lengthening of dry spells.
The duration and probability of occurrence of the other extreme events (hot spells and strong winds) in the rainy season are very high in the study area at the beginning of the rainy season.
The results presented in this study show that rainy season potentialities and constraints could be understood from the daily climate information. Thus, the prediction of the different characteristics of the rainy season (annual rainfall amount, onset and cessation of the rainy season, dry spells, heavy rain events, hot spells, and strong winds) in the region from climate models would help local farmers plan and adequately organize their field activities. Such strategies, developed through national frameworks would then ensure food security at the local level, but most importantly address issues of crop failure across Sahelian regions.