Journal of
Hydrology ELSEVIER
Journal of Hydrology 188-189 (1997) 74-96
Rainfall monitoring during HAPEX-Sahel. 1. General rainfall conditions and climatology T. L e b e l a'*, J.D. T a u p i n b, N. D ' A m a t o b aORSTOM/LTHE, Groupe PRAO, BP 53, 38041 Grenoble Cedex 9, France bORSTOM/LTHE, Groupe PRAO, BP 11416, Niamey, Niger
Abstract The HAPEX-Sahel experiment took place in the midst of the most severe drought that has ever plagued the region since rainfall records have been available in the Sahel. The aim of this paper
T. Lebel et aL/Journal of Hydrology 188-189 (1997) 74-96
75
associated EPSAT-Niger (Estimation des Pluies par Satellite, exprrience Niger; Lebel et al., 1992) experiment, provided the opportunity to look more closely at the inter- and intra-annual variability of rainfall over a region that was harshly hit by the drought. The installation of the EPSAT-Niger (E-N) setup began in 1989 in order to provide a comprehensive rainfall coverage of the HAPEX-Sahel (H-S) study area from 1990. It is briefly described in Section 2, a more complete description being available in Lebel et al. (1995). The general rainfall conditions that were observed during the HAPEX-Sahel long-term monitoring period (LTP; 1990-1993) are then described in Section 3 and compared with the long-term rainfall statistics computed for the area before and during the drought. The results of the climatology study of Le Barb6 and Lebel (1997) are used to provide reference values with which the H-S statistics can be compared. This analysis clearly shows that the LTP years prolong the dry sequence which began at the end of the 60s. A more detailed analysis of the rainfall climatology at the event scale is then presented in Section 4. Point and area averaged rainfall statistics are computed for both the nonconditional and conditional to non-zero rainfall process. The year to year fluctuations of these statistics are analysed. The dynamics of the rainfall events is also studied. In Section 5, rain rates are considered. Although tipping bucket raingauges provide time integrated raindepths, with an integration time which depends on the rainfall intensity, representative rain rate distribution curves are established. Concluding remarks regarding the climatology of the Sahelian rainfall are given in Section 6, introducing the companion paper by Lebel and Le Barb6 (1997).
2. Rainfall measurements during HAPEX-Sahel, using the EPSAT-Niger setup The EPSAT-Niger setup combined, for the first time in West Africa, a digitised weather radar and a network of 100 recording raingauges, covering an area of 16 000 km 2 (Lebel et al., 1992). The radar data are currently being processed. They are not yet usable for computing reliable quantitative rainfall estimates, mainly because accounting for the attenuation affecting C-band measurements requires care due to numerical divergences in the correction algorithms (Lecocq et al., 1994; Benichou et al., 1995). Nevertheless the radar data proved useful for the analysis of the dynamics of the Sahelian mesoscale convective systems (SMCSs) presented in Section 4.4 below. The preliminary study, carded out in 1989, covered exactly the H-S 1 x 1° square (DS), bounded by latitudes 13°N-14°N and longitudes 2°E-3°E. However, as this zone coincides with less than 25% of the working area of the Niamey weather radar, it was decided in 1990 to extend the network so that it covered a 16 000-km 2 rectangle between latitudes 13°N-14°N and longitudes l°40E-3°E (Fig. 1). 'The network pattern over the DS is a roughly regular grid with nodes spaced at 12.5 kin. A target area (TA) of 875 km 2 including the East Central Super Site (ECSS) and the West Central Super Site (WCSS) was equipped with 29 gauges (Fig. 1). A smaller TA (six gauges over 100 km 2) was also created to cover the Southern Super Site (SSS). A complete description of the E-N setup may be found in Lebel et al. (1995); the H-S experimental setup and the rationale of the super sites are presented in Goutorbe et al. (1994). In the following we shall distinguish the basic network covering the DS stricto sensu with a regular grid (around 70 stations from
T. Lebel et aLlJournal of Hydrology 188-189 (1997) 74-96
76
EPSAT-Niger: 1992 raingauge network (107 sites). -40-30-20-10 0 10 20 30 40 50 60 70 80 90 100 110 i l l / , - , ' - ] I ,~4 f I I I I I I • .33 =36
-~ 90
iPo
.78
.45
.77
i122
.~
70
120 110
90
i~
4~leyor °
.~
80
110 I
.4~
,*o
80
.71
,~
,61
70
~m 866
40
.91
=~
87
b31
30
,17;
.~22
&59
•5§
40
,6? =84
F
20
~ 30
.39
.51
.s
10
-
,62
A4,1
.44
6~
"~
0 -40-30-20-10
0 10 .20 30 Coordinotes in km -
~Birnin G,
--
40 50 6 0 70 Origin: 2 E; 13 N
80
90
100
0 120
110
: 30 stations
60
70
80
I
I
I
m114
,111 70
70 .114
.121
,lO~m
.t10
60
~i
,r/,~ m .13
60
,loo ,tll
~5 T20 ¸
.23
812 50 50
I 60
10
.32
Network over the central supi~mitss
50
!
.s2
e ~s •
.57
I 70
I 80
50
77
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
Table 1 Proportionof the annual rain depth falling during the rainy season. The first figure is the rainfall in ram, while the second is in % of annual rainfall.The values given for the five EPSAT-Niger stations are averages over 19911993
Year 15/4-15/10 1/5-1/10
Niamey 1905-1989
Niamey Atro.
564 556 539
480 479 459
100 99 95
100 100 95
Niamey ORSTOM
Banizoum (ECSS)
Fandou (WCSS)
IH Falow (SSS)
501 497 485
462 453 440
506 506 492
626 619 619
100 99 97
100 98 95
100 100 97
100 99 99
IH, Institute of Hydrology.
1990 to 1992), from the total network which, in addition, includes the T A stations and the ten stations installed to the west of the DS. The gauges used are of a classical tipping bucket type, with a 400-cm 2 collector. Each tipping, corresponding to 0.5 m m of rainfall, is recorded to the second on a read only memory. Intensities of up to 1800 m m h -1 are thus measurable. The entire network operated permanently between mid-April and the end of November each year. A dozen stations operated all year long. On two sites the rainfall measured at a height of 1.5 m above ground (standard height for the E-N network) was compared with the rainfall measured on the ground. N o significant differences were found between those two measurements. The rain water is collected into a bottle after the bucket tipping, allowing the correction of the cumulative rainfall when the tipping does not correspond exactly to 0.5 mm of rain. All the statistics given below for time steps longer than 1 day are based on those bottle-corrected rain measurements.
3. General rainfall conditions To describe the rainy conditions that prevailed during H-S, in particular with respect to the climatology of the recent drought, the 6-month period between the 10th of April and the 10th of October is considered. As may be seen from Table 1, during this period falls 99% of the annual rain, both on the 1 9 0 5 - 1 9 8 9 Niamey series and during H-S. A shorter period of 5 months (1/05-1/10) still accounts for more than 95% of the annual rainfall during the H-S years, in accordance with the long term statistics. It is thus observed that the length of the rainy season over the period 1990-1993 conforms to the average. 3.1. Drought continued
78
T. Lebe l et aLIJournal of Hydrology 188-189 (1997) 74-96
Table 2 Statistics of the point rainfall values recorded by the basic network stations of the DS over the rainy season (10/410/10). The number of stations is smaller in 1991 due to a late installation of the network. In 1993 the basic network was reduced to allow for a denser instrumentation of the target area Year
No. of stations
Mean (rap)
SD ($)
Mini. (m)
Maxi. (M)
CV (%) (s/mp)
(M - m)/ mp (%)
Niamey Niamey ORSTOM A6ro.
396 523 513 459
63 95 71 83.9
292 341 389 315
659 725 782 622
16,0 18.2 13,9 18.6
93 74 77 65
399 541 488 447
(n) 1990 1991 1992 1993
60 36 70 30
474 434 607 399
over the period of rainfall record availability. This drought has been documented by several authors (Nicholson, 1980, 1981; Lamb, 1982, 1983; Hubert and Carbonnel, 1987) and a recent analysis centred on the H-S study area is presented by Le Barb6 and Lebel (1997). The rainfall climatology during H-S will be globally characterised by comparing some E-N seasonal rainfall statistics with those of the only long-term rainfall series (Niamey) available in the H-S square. The Niamey rainfall statistics being very similar over the years 1905-1989 and over the years 1950-1989 (Le Barb6 and Lebel, 1997), this latter period will serve to define the 'standard' rainfall climatology of the Niamey area. The period 1950-1967 will be used as reference for extremely wet conditions and the period 1968-1989 as reference for extremely dry conditions (the rainfall index computed for the whole Niger by Le Barb6 and Lebel, 1997, indicates that 1968 was the first year of a period of continuous rainfall deficit). In Tables 2-4, various measures of the precipitation over the DS are presented: (i) the average of the point values recorded by the basic network (the over-sampling Of the super sites is eliminated; this average is denoted mp); (ii) the average of the point values recorded by all the available stations (the over-sampling of the super sites is kept; this average is denoted mRs); (iii) the area average over the DS, computed by optimally interpolating (kriging) the whole set of available data, denoted/~ds. Generally m v and #ds are not significantly different (Table 4), which was to be expected, given the regular and dense sampling associated with the basic network. With respect to the dry period, 1991 and 1992 were above normal (Fig. 2). In fact, for both years, m v is only 5% smaller than the median of the 1950-1989 annual series. On the Table 3 Statistics of two EPSAT-Niger space-time series and of three time series of the Niamey A6roport station Sample
Size (n)
Mean
SD (s)
(rap) 9 gauges90-93 !1 g a u g e s 9 0 - 9 3 Niamey 50-89 Niamey 50-67 Niamey 68-89
33 40 40 18 22
465 457 564 654 495
85 86 123 145 108
Mini.
Maxi.
CV (%)
(M - m)l
(m)
(M)
(s/mp)
mp (%)
312 312 294 454 294
622 622 980 940 689
18.4 18.9 22 22 22
67 68 114 74 80
Median
477 461 549 627 499
T. Lebel et aL/Journal of Hydrology 188-189 (1997) 74-96
79
Table 4 Area-averaged seasonal rainfall (10/4-10/10) over the DS and the super sites, computed by kriging of the point values. Areal estimates are given with their standard deviation of estimation error. The point value recorded at Banizoumbou is given to illustrate the spatial averaging effect mp (basic net.)
rn as # ds SSS (all stations) (12000 km 2) (100
1990 1991 1992
396 523 513
405(74) 524(52) 513(98)
396 + 2 522 ± 4 511 +-- 3
1993
459
477(99)
460
Mean
473
480
472
419 627 603 501 538
kln 2) --. 23 ± 19 ± 10 + 6
WCSS (225 km 2)
ECSS (400 km 2)
Banizoum. (ECSS)
386 537 493 470 481
385 - 15 560 ± 14 500 ± 6 458 479
402 494 410
± ± ±
15 16 14 3
441
other hand, in the years 1990 and 1993, mp is smaller than the mean and the median of the 1968-1989 series. The year 1990 was especially dry with one station recording only 292 mm of rainfall, that is less than the absolute minimum of the 1950-1989 Niamey series. The 4 H-S years averaged over the DS is equal to 472 mm, that is 5% below the average of the dry period, and 16% below the 1950-1989 average. The situation observed over the H-S study area was in agreement with that of the whole of Niger, 1990 being generally dry, 1991 and 1992 being wetter (Lebel et al., 1995). The drought observed since 1968 has thus continued during the H-S years. 3.2. Time and space distribution at the seasonal scale
The time distribution within the rainy season was not identical each year. Except for 1990, a spell of reduced rainfall was recorded each year during the core of the rainy
=
DS average
&
Minimum
--o
Maximum o
|
Niamey
(~STOM
==
Niamey Aero.
J
1990
1991
1992
. . . . .
1950-67
.........
1968-89
1993
Fig. 2. Yearly rainfall statistics of the EPSAT-Niger data set, compared with the inter-annual averages of the wet (1950-1967) and dry period (1968-1989), respectively.
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
80
Area
averaged
rainfall
over
ECSS
1,0
0,8 C
(D >
0,6
E o
0,4
"O qD 0 U)
0,2
OsO
m 0
m 0
~, 0
~ 0
o 0
¢.0
~ 0
~,0
~ 0
w 0
m 0
o 0
m 0
m 0
o ~"
10 day periods Fig. 3. Intra-seasonal distribution of the ECSS rainfall for the 4 HAPEX-Sahcl yeats (see Table 4 for a comparison between the ECSS and DS yearly rainfall). The 10-day rainfall is scaled by the seasonal total in order to compare the rainfall chronologies. The Niamey series (1905-1989) is used as a climatologic reference. season. This is clearly visible in Fig. 3, where the intra-seasonal rainfall distribution is characterised by scaling the 10-day rainfall by the seasonal rainfall:
[(k) =Z(k)/Zs where Z(k) is the cumulated
(1)
rainfall over the kth 10-day period of the season and Zs is the seasonal rainfall. Z may be either a point or an area averaged rainfall. In mid-June the mean position of the l(k) curve for the 4 H-S years is centred on the 1905-1989 curve. From there to mid-September the yearly H-S curves (except 1990) are permanently below the 1905-1989 curve (in 1991 the deficit starts in mid-July only). Since this profile is computed for the ECSS average rainfall, this departure from the long-term average cannot be attributed to a local sampling effect, as could be the case for an isolated station. This behaviour is in agreement with the finding of Le Barb6 and Lebel (1997) that the ongoing dry spell is mostly characterised by a reduced rainfall during the core of the rainy season.
The 1992 intra-seasonal distribution of rainfall was the most irregular of all. A m a x i m u m deficit of nearly 50% (as compared with the 1968-1989 average) was reached at mid-July, causing an important delay in the vegetation growth that was never totally
81
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96 ECSS r(]infoll(mm): 1 5 / 0 4 - 1 5 / 1 0 66 76
I__--"
,I
/
1992 86
WCSS: 1 5 / 0 4 - 1 5 / 1 0 52 62
I
/
1992
ol
/ uJ
20
I
0" 0
1
,
l
2
,
I
,
3
I
4
,
I
5
,
6
-(Ln(1-F)) Fig. 6. Distribution of the event rainfall over the HAPEX-Sahel 1 x 1° square and at the Banizoumbou station (East Central Super Site). An exponential model (F(r) = 1 - exp(rls)) has been fitted to each distribution. Except for the four largest values the Banizoumbou fit is good, while the ! ° x 1° square model deviates markedly from the observed distribution when F is over 0.9.
denoting H as the point cumulative event rainfall, the conditional (H > 0) cumulative distribution function (c.d.f.), F, is given by: F(H)
(2)
= 1 - e (-n/s)
where s is both the mean and the standard deviation of the distribution. Fig. 6 shows the example of Banizoumbou where the exponential distribution fits well the observed data (N = 117, mH = 13.2 and sH = 13.8), whereas a significant departure is observed for the DS distribution (N = 172, roDS = 10.2 and s o s = 8.5). Logically the area averaged rainfall has a smaller coefficient of variation. The spatial averaging smoothes out the large point values and eliminates the zeros, which diminishes the standard deviation, while the average remains constant. A gamma distribution gives the best fit to the observed series of DS averaged event raindepths, denoted Has. The probability density function of Has is:
Has
/2Le-s f(nds) =
sXi,(h )
(3)
with s, scale parameter, equal to 7.4, and X, shape parameter, equal to 1.4. In order to assess the inter-annual fluctuations of the point event rainfall distribution, the exponential model was also fitted on a yearly basis. When the estimate, s*, of the exponential distribution parameter s is the sample mean, its standard deviation is
86
T. Lebel et aL/Journal of Hydrology 188-189 (1997) 74-96 Banizoumbou 80
.
1990-1992
.
.
a
Observations 1990
o
Observations 1992
.
,_. E 6O
".........
model 90, s=15.9
..-~3
E
~
...............
model 91, s=14.7 model 92, s=12.8
. ..-'"'°~/" .-'" ~
~'~
~
._'~
oo°
[]
~
?j" °°°°°
[] ~
°.°
i'-i °.o°°°
.....
...,." O
.................. . .-""" 0
....."0
......"'~
o >
20
UJ
f
• °°
..o'~""
,..s °
0 ~ 0,0
"
1,0
I 2,0
,
I 3,0
,
I 4,0
-(LN(1-F))
Fig. 7. Event rainfall distributions for the years 1990, 1991 and 1992 at Banizoumbou. The observations for 1991 are not plotted for clarity of the figure. The model is an exponential distribution.
given by:
a(s*) =s*/ ~n
(4)
where" n is the sample size. Thus, even for Banizoumbou, which was one of the stations where the rainfall statistics fluctuated the most from year to year, with a particularly low mean event rainfall of 11.8 mm in 1992, the mean event rainfall computed each year remains in the 1 standard deviation confidence interval. That year, with n = 32, a(s*) is equal to 2.6 mm. The lower bound of the 1 standard deviation confidence interval is then 10.6 mm if considering the overall Banizoumbou average a n d l l . 1 mm if considering the 30 test stations' average (13.7 mm). Furthermore, even with such a low mean event rainfall the distribution of the 32 observed values remained close to an exponential (Fig. 7). At that point, several conclusions regarding the event rainfall climatology may be drawn. Firstly, the point event rainfall, conditional to non-zero values, may be assumed to be approximately exponentially distributed with an average of around 14.0 mm:
F(H) = 1 -
e (-H/14)
(5)
Secondly, the probability of zero rainfall in a SMCS is close to 1/4. The accuracy of this figure is conditioned by the E-N observation window being of an appropriate size, a fact that seems verified by the radar observations. Finally, since the inter-annual fluctuations of the mean event rainfall are weak, it is essentially the variation of the number of events observed each year that explains most of
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
87
the inter-annual rainfall variability. Note that Points i and 3 are in total agreement with the findings of Le Barb6 and Lebel (1997), working with a different approach on a different data set (long-term series of the Central Sahel). 4.3. Classification o f the rainy events
Amani et al. (1996) have proposed a classification of the Sahelian event rainfields based on a statistical parameter (UPA) characterising the spatial structure of the rainfield. A first group of events seems to correspond to the well-structured squall lines, with a very low probability of zero rainfall (less than 5%). A second group is composed of the SMCSs characterised by a higher probability of zero rainfall (F0 = 26%) and a smaller average point rainfall (13.7 mm, against 22.2 mm for the events of Group 1). The last group is characterised by a still higher probability of zero rainfall (F0 = 41%) and a smaller point rainfall (9.7 mm). It may be associated with localised convective rain. Since the events of these groups have both a different spatial structure and different point statistics, the question is raised of whether they should be considered as belonging to different point processes. The event rainfalls of each group are plotted in Fig. 8 for Banizoumbou. Generally the events of Group 1 produce larger rain than those of Group 2 and those of Group 2 larger rain than those of Group 3. Nevertheless the two heaviest rains at Banizoumbou were recorded for events of Group 2. This group is the only one for which the exponential distribution holds. Further more its parameter (14.2 mm) is very close to that of the full Banizoumbou 1990-1993 sample (13.2 mm, Fig. 6). There is no clear indication at this stage that the point processes of the three groups should be distinguished. There is
Banizoumbou 80
i
• Obs,Group 1 • Obs,Group2 a Obs.Group 3 "
6o
1990-1993
i
•
"
1
mm
mm/~m
m m m m~m,~~/~ uJ 20 ~
0
. . . .
m
l
mmam• a
I
m
I
i
[ ]
Model Group2, s=14.2 •
i
•
a
[]
a
I
I
L
I
2
3 -LN(1
I
I
4
-F)
Fig. 8. Eventrainfalldistributionsfor the three UPA groupsdefinedby Amaniet al. (1996). An exponentialmodel has been fitted to the data of Group 2.
88
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
obviously a link between the degree of intermittency of a rainy event and the average point rainfall it produces: more intermittent rainy events tend to produce less rainfall where it rains. Nevertheless, since the exponential model fits well the whole point event rainfall sample for the subset of 30 test stations, the hypothesis of a global exponential distribution of the event rainfall point process will be maintained. On the other hand, the results of this classification should be taken into account for estimation problems, since high values of F0 preclude the use of classical interpolation methods for the events of Group 3. This will be further analysed in Lebel and Le Barb~ (1997). Also it provides a simple diagnostic tool to compare the nature of the rainy events observed from one year to another. Amani et al. (1996) have observed that the proportion of each group has been relatively stable for 3 of the 4 H-S years, but that a significantly smaller number of Group 1 events were recorded in 1992 (10% against 21% in 1991). It will thus have to be investigated whether, in addition to the total number of rainy events observed during a rainy season, the proportion of each group is another factor to take into account when looking for possible changes in the rainfall climatology.
4.4. Dynamics The statistics presented above are Eulerian in nature: the convective systems are observed while passing over a given location and the total rainfall it produces is recorded. Several applications require to know more than these Eulerian statistics. For instance, when estimating rainfall at short time steps, time correlation is equally important as spatial correlation. According to the Taylor hypothesis, the time correlation and spatial correlation functions are related to each other through the velocity vector when moving rainfall systems are considered. The radar data collected during EPSAT-Niger, while not yet usable for quantitative purposes, have the great advantage over raingauge data to allow for the survey of an entire convective system. Forty-five of the mature SMCSs that passed over the H-S square during the three rainy seasons 1990-1992 were successfully monitored from their entry into the east of the area covered by the Niamey weather radar, until the convective edge left the western part of the radar influence zone, 6 to 8 h later. Average speeds and moving directions have been estimated by computing the distance covered by the edge of the SMCS between two consecutive images. The time interval between two images ranges from 5 min (when the SMCS is right over the E-N target area) to 20 min (when the convective part of the SMCS is outside the H-S square). The histograms of the speed and moving direction are shown in Figs 9 and 10. The average speed is of the order of 55-60 km h -1 and 75% of the observed SMCSs moved at a speed ranging from 45 to 70 km h-l. A distinction has to be made between the SMCSs moving at a constant speed during the period of observation (40% of the sample) and those whose speed fluctuated (50%, the remaining 10% being declining SMCSs, whose speed decreased steadily during their passage over the H-S square). The SMCSs of the first category had a slightly higher average speed (61 km h -i) than the SMCSs of the second category (54 km h-l). A similar proportion of SMCSs displayed a constant (45%) and fluctuating (50%) moving direction. Given the imprecision in their computing, differences between average moving directions of the first (245 °) and second (250 °) category are probably not significant. The steadiness of the speed and direction of displacement of the SMCSs
89
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96 Sahellan
Mesoscsle
Convective
Systems
1990-1992
111
•:)
6
@ "6
~,¢s
4
E .-s
Z
0 40
50
60
Averaged
speed
70
80
(kmlh)
Fig. 9. Average moving speeds for 45 Sahelian mesoscale convective systems observed over the area covered by the Niamey weather radar. Duration of observation is of the order of 6 to
90
T. Lebel et aL/Journal of Hydrology 188-189 (1997) 74-96
led Amani and Lebel (1997) to propose a Lagrangian approach for estimating Sahelian rainfall at short time steps.
5. Rain rate distribution
The analysis of the rain rate distribution is of interest to the study of several problems. The development of climatologic and thresholding algorithms for rainfall estimation from either radar or satellite data is one of those (see e.g. Kedem et al., 1990). A fact that has to be noted, with respect to remote sensing of rainfall, is that microwave radiometers are often saturated for rainfall intensities of a few tens of mm h-t. Knowing the proportion of rain that falls at intensities higher than any given threshold may thus help in the design of satellite rainfall measurement missions such as Tropical Rainfall Measurement Mission (TRMM) (Simpson et al., 1988) and TROPIQUES (Desbois, 1994). The analysis of rain rate distributions is also important for the modelling of the rainfall time-variability. The main difficulty of rain rate analysis is linked to the measurements. Rainfall is a discrete process made of droplets. To obtain meaningful rates some time integration is needed. Tipping bucket gauges integrate the rainfall signal over constant height increments (0.5 mm in the case of the E-N gauges), the time step of integration depending on the rainfall intensity. For high rain rates the time between two consecutive tippings is less than I rain (60 s for an average rain rate of 30 mm h -t, 18 seconds for an average rain rate of 100 mm h-t). On the other hand it takes 5 min between two tippings for an average rain rate of 6 m m h -1. It is consequently difficult to find an optimal time step to convert the series of durations between tippings into a constant time increment series. A short time step (of the order of 1 rain or less) requires a time interpolation at low intensities, while longer time steps (of the order of 5 min or more) integrate and smooth out high intensities, biasing the upper part of the rain rate distribution.
Rain
rate
distributions
1990
i e/ •
40-
>
.
3 E 0
20"
-i
0
"
0
50 Rain
J
100 Rates
"
|
150
•
200
(mrn/h)
Fig. 11. Rain rate distributions for an array of stations in 1990.
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
91
Here, the inter-tipping durations recorded by the E-N stations have been directly converted into average rain rates. The annual c.d.f.s of these average rain rates have then been computed for each station from 1990 to 1993. The array of c.d.f.s obtained each year is bounded by two envelope curves as shown in Fig. 11 for the year 1990. That year, the lower bound is given by the Tanaberi (N°32 in Fig. 1) c.d.f., while the upper bound is given by the Sandideye (N°57 in Fig. 1) c.d.f. The median rain rate is 50 mm h -1 in Tanaberi which recorded the largest seasonal total in 1990 (660 mm) and 23 mm h -1 in Sandideye, which had one of the smallest seasonal totals (353 mm). Rain rate distributions observed each year are compared in Fig. 12 (envelope curves) and in Fig. 13 (average distributions). In both figures, distributions are similar between years, especially in Fig. 13. The overall median rain rate (50% of the seasonal precipitated water) is around 35 mm h -1 and 1/3 of the total precipitation fails with an intensity higher than 50 mm h -l. Such results seem reliable. The bias resulting from the assumption that the rain rates are constant during the interval between two consecutive bucket tippings is likely small for inter-tipping intervals shorter than 1 rain (rain rates above 30 mm h -1) which means that the computed distribution for rain rates above 30 mm h -t is little affected. A certain confidence is thus attached to the estimation of the median (35 mm h-i), as well as to that of the upper quartile (75 mm h-t), and these numbers are of great importance in the light of the remote sensing needs enunciated above. Another striking fact when analysing the envelope curves of Fig. 12 is that, each year, the upper distributions (low probability of high rain rates) all come from low seasonal rainfall stations, whereas the lower distributions (higher probability of high rain rates) all come from high seasonal rainfall stations. This calls for two main comments.
~
oo
~ 8o
P
~ 40 i I lff t[l'f
• ,
I ~l" ~ 8 20 L I r J l I r~r"
E
~
0 0
20
40
60
80
T...l>,,,i,o
i,
Gankiba 91
; • i
Nlabere 92 Niamey lero 92 N_iameyAero. 93 "l'anaberi93
100
120
140
160
180
200
Rain Rates (mm/h) Fig. 12. The extreme rain rate distributions observed for each of the 4 HAPEX-Sahel years.
92
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
~ 00
'
|
•
!
•
|
•
I
•
i
•
|
•
!
•
g. o IL
~r~4f d~
s A •
E(~20 ~
H
0
O
Average1991 Average1992 Average1993
I
i
I
20
40
i
i
i
i
,
I
i
i
l
,
60 80 100 120 140 160 190 200 Rain Rates (ram/h)
Fig. 13. Average rain rate distributions for each of the 4 HAPEX-Sahel years.
First, as long as a sufficient number of rainy events are observed in a rainy season, a representative sample is obtained and the rain rate distribution does not change much between dry and wet years. Second, the differences between rain rate distributions recorded at different stations for a given rainy season probably explain most of the strong inter-station seasonal rainfall variability. Indeed, the number of rainy events may vary dramatically from one year to another, thus explaining most of the overall inter-annual rainfall fluctuations, but this variation is not so important in space over a limited range of 100 km: it cannot account for large seasonal rainfall differences between stations located a few kilometres apart. Rather, the granular structure of the event rainfall, linked to the convective cells, should be considered to explain these local gradients. Stations recording over-the-average seasonal rainfall are those frequently hit by mature convective cells producing high rain rates (lower curves in Figs. 11 and 12). This would mean that the average number of rainy events observed over 1 year (say, between 35 and 50) is sufficient to provide a representative sample of rainy events so that their time accumulation smoothes out mean rainfall event statistics. But a much larger number of events is required to smooth out the spatial variability associated with the convective cells. In Taupin et al. (1993b) an isohyetal map of the rainfall cumulated over the 3 years 1990-1992 is shown. It still displays strong local gradients. The ongoing long-term rainfall monitoring programme over the H-S square will provide the necessary data to assess the minimum number of events required to obtain east-west oriented isohyets, as usually visible on the long-term averaged rainfall maps of West Africa. It should not be excluded either that there is a non-random component, linked to the evaporation physics, driving the succession of convective cells at a given location
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
Banlzoumbou
93
1992
100 .........m-
J
I
80
O . w
.Q
60
z.
>
40
o D
m
E (.,)
20
0
I
0
I
10
20
Effective
1
I
30
Rainfall
40
Time
I
50
60
(hours)
Fig. 14~Effectiverainfall duration at Banizoumbou. during the rainy season, a possibility that could be thoroughly investigated thanks to the very comprehensive H-S data set. Another statistic that is computable from the series of inter-tipping intervals is the effective duration of rainfall. This computation assumes that there was no interruption of rainfall between two consecutive bucket tippings. Such an assumption cannot hold for inter-tipping intervals longer than 10 min, but is certainly reasonable for inter-tipping intervals shorter than 5 min, which account for more than 80% of the EPSAT-Niger records. The ranking of inter-tipping intervals in increasing order thus allows the drawing of a curve of effective duration of rainfall that is valid up to at least the 80% limit. An example is given in Fig. 14 for Banizoumbou, showing that more than 50% of the rain effectively falls in less than 5 h, underlying the intermittency of the Sahelian precipitation.
6. Concluding remarks This paper was intended to provide an overview of the rainfall conditions during HAPEX-Sahel (H-S). The results presented here come from the analysis of the EPSATNiger raingauge data set, that is 100 recording raingauges covering the HAPEX-Sahel square and operated over the years 1990-1993.
94
T. Lebel et aL/Journal of Hydrology 188-189 (1997) 74-96
The H-S years have seen the continuation of the drought that has struck the Sahel for 25 years now. The years 1991 and 1992 were moderately dry over the H-S square, whereas 1993 and, above all, 1990 were drier than the 1968-1989 average. This global precipitation deficit was associated with a great irregularity in the space and time distribution of rainfall, which appears to be a much more intermittent process than usually believed, even after integration over large time spans. In 1992, the feature of greatest significance was that the pre-IOP period was unusually dry, even with respect to the statistics of the ongoing drought, and that much heavier rainfall than normal was then recorded during the lOP. Also, based on the classification proposed by Amani et al. (1996), a smaller proportion of extensive mesoscale convective systems were observed that year. Hence, by many different standards, the rainfall climatology of the lOP year appears to have been very atypical. However, by contrast with the strong variability of the seasonal rainfall, from year to year and from station to station, some elements of the rainfall climatology observed over the H-S square were found to be remarkably stable. First of all, the length of the rainy season was similar for all 4 years, indicating that the seasonal precipitation amount is not primarily linked to that parameter. The event rainfall statistics are also very stable from one year to another. The point event rainfall is exponentially distributed (i.e. at a given station, the c.d.f, of the event rainfall is an exponential), with an average of around 14.0 mm. When the occurrences of zero rainfall are taken into consideration (i.e. a rainy event was observed over the H-S square but it did not produce any rain at the station considered), the average decreases to 10.5 mm. A similar figure is obtained when computing the mean yield of the rainy events over the H-S square, implying the first order stationarity of the rainfall process (the point and areal averages are equal) and a probability of zero rainfall equal to 1/4. In contrast with the inter-annual stability of the mean event rainfall, the annual number of events vary in close proportion to the annual rainfall (39 events for 396 mm in 1990; 47 events for 522 mm in 1991; 48 events for 511 mm in 1992). It may thus be concluded that the seasonal rainfall deficit is more due to a smaller number of rainy events than to a variation of the magnitude of these events. Looking at a smaller scale, the rain rate distributions also appeared to be fairly constant, allowing the computation of an average distribution for each year. These distributions are surprisingly close to each other. Half the annual rain falls at intensities exceeding 35 mm h -~ and 1/3 at intensities higher than 50 mm h -1. The effective time of precipitation is accordingly small, with 50% of the seasonal precipitation recorded at any given point falling in less than 5 h. Such an intermittency makes Sahelian rainfield sampling especially difficult, whether on the ground or by remote sensors. In addition, the high frequency of intensities exceeding 35 mm h -1 may prove to be a significant impediment in the measurement of rainfall by microwave radiometers, given their saturation limits. Whether looking at seasonal or event rainfields, the granular structure of the Sahelian mesoscale convective systems, composed of convective cells, causes a high spatial rainfall gradient, which encourages a degree of caution when interpolating these fields. In the companion paper by Lebel and Le Barb6 (1997), it will be examined to what extent sufficiently accurate estimates of the areal event rainfall may be obtained to meet the requirements of climate and hydrologic modellers.
T. Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
95
Acknowledgements This work is the result of the large efforts that were put in by the EPSAT-Niger team during 4 years spent in Niger. Nothing would have been possible without the collaboration of the Niger Direction de la M6t6orologie Nationale (DMN). The contributions of Jean Lecocq in writing the code used for the rain rate distribution section and of Dr Baxter Vieux in carefully checking the manuscript are gratefully acknowledged. This research was funded by the French Ministry for Cooperation and ORSTOM. References Amani, A. and Lebel, T., 1997. Lagrangian kriging for the estimation of Sahelian rainfall at small time steps. J. Hydrol., in press. Amani, A., Lebel, T., Taupin, J.D. and Rousselle, J., 1996. Typology of rainfall fields to improve rainfall estimation in the Sahel by the ATI method. Water Resour. Res., 32(8): 2473-2487. Arnaud, Y., Desbois, M. and Maizi, J., 1992. Automatic tracking and characterization of African convective systems on Meteosat pictures. J. Appl. Meteor.; 31: 443-453. Benichou, H., Delrieux, G. and Lecocq, J., 1995. Rainfall climatology in tropical Africa using a C-band weather radar: (1) the attenuation problem. In: Proceedings of the Third IAHR International Symposium on Hydrological Applications of Weather Radar, August 1995, Sao-Paulo, pp. 361-374. Desbois, M., 1994. TROPIQUES, a small satellite for the study of the variability of water and energy cycles in the intertropical band. In: Proceedings of the European Symposium on Satellite Remote Sensing, 26-30 September 1994, Rome. SPIE-EUROPTO series, Paper 2317-14. Desconnets, J.C., Taupin, J.D., Lebel, T. and Leduc, C., 1996. Hydrology of the HAPEX-Sahel Central SuperSite: surface water drainage and aquifer recharge through the pool systems. J. Hydrol., this issue. Eagleson, P.S., 1972. Dynamics of flood frequency. Water Resour. Res., 8(4): 878-898. Goutorbe, J.P., Lebel, T., Tinga, A., Bessemoulin, P., Brouwer, J., Dolman, H., Engman, E.T., Gash, J.H.C., Hoepffner, M., Kabat, P., Kerr, Y.H., Monteny, B., Prince, S., Said, F., Sellers, P. and Wallace, J., 1994. HAPEX-SAHEL: a large-scale study of land-atmosphere interactions in the semi-arid tropics. Ann. Geophys., 12: 53-64. Hubert, P. and Carbonnel, J.P., 1987. Approche statistique de l'aridification de I'Afrique de I'Ouest. J. Hydrol., 95: 165-183. Kedem, B., Short, D.A. and Karni, Z., 1990. An analysis of the threshold method for measuring area-average rainfall. J. Appl. Meteor., 29: 3-20. Lamb, P.J., 1982. Persistence of subsaharan drought. Nature, 299: 198-212. Lamb, P.J., 1983. Subsaharan rainfall update for 1982. Continued drought. J. Clim., 3: 419-422. Le Barb6, L. and Lebel, T., 1997. Rainfall climatology of the HAPEX-Sahel region during the years 1950-1990. J. Hydrol., this issue. Lebel, T. and Le Barb~, L., 1997. Rainfall monitoring during HAPEX-Sahel. 2. Point and areal estimation at the event and seasonal scales. J. Hydrol., this issue. Lebel, T., Sauvageot, H., Hoepffner, M., Desbois, M., GuiUot, B. and Hubert, P., 1992. Rainfall estimation in the Sahel: the EPSAT-NIGER experiment. Hydrol. Sci. J., 37(3): 201-215. Lebel, T., Taupin, J.D. and Gr6ard, M., 1995. Rainfall monitoring: the EPSAT-Niger set-up and its use for HapexSahel. In: T. Lebel (Editor), Hydrologie et M6t~orologie de M6so-Echelle dans HAPEX-Sahel: Dispositif de Mesures au Sol et Premiers R6sultats. 31-68, ORSTOM Lecocq, J., D'Amato, N., Cazenave, F. and Lebel, T., 1994. R6sultats pr61iminaires de la campagne EPSAT-Niger pour la calibration d'un radar bande C. In: M. Hoepffner, T. Lebel and B. Monteny (Editors), Interactions Surface Continentale/Atmosph~re: L'exp6rience HAPEX-Sahel, 10 ~me Journ6es Hydrologiques de I'ORSTOM, Montpeilier. ORSTOM, Paris, pp. 547-562. Nicholson, S.E., 1980. The nature of rainfall fluctuations in subtropical West Africa. Mon. Wea. Rev., 108: 473486.
96
1".Lebel et al./Journal of Hydrology 188-189 (1997) 74-96
Nicholson, S.E., 1981. Rainfall and atmospheric circulation during drought periods and wetter years in West Africa. Mort. Wea. Rev., 109: 2191-2208. Roweil, D.P. and Millford, J.R., 1993. On the generation, of African squall lines. J. Clim., 6: 1181-1193. Simpson, J., Adler, R.F. and North, G.R., 1988. A proposed Tropical Measuring Mission (TRMM) satellite. Bull. Am. Met. Soc., 69: 278-295. Smith, R.E. and Schreiber, H.A., 1974. Point process of seasonal thunderstorm rainfall 2. Rainfall depth probabilities. Water Resour. Res., 10(3): 418-423. Taupin, J.D, Amani, A. and Lebel, T., 1993a. Small scale spatial variability of the annual rainfall in the Sahel. In: H.-J. Bolle, R.A. Feddes and J. Kalma (Editors), Exchange Processes at the Land Surface for a Range of Space and Time Scales, Proceedings of the Yokohama Symposium, July 1993. IAHS Publ. No. 212, pp. 593-602. Taupin, J.D., Lebel, T., Cazenave, F., Greard, M., Kong, J., Lecocq, J., Adamson, M., D'Amato, N. and Ben Mohamed, A., 1993b. EPSAT-NIGER, campagne 1992. Rapport ORSTOM-DMN, 98 pp.
