15 OCTOBER 2006

BELL AND LAMB

5343

Integration of Weather System Variability to Multidecadal Regional Climate Change: The West African Sudan–Sahel Zone, 1951–98 MICHAEL A. BELL*

AND

PETER J. LAMB

Cooperative Institute for Mesoscale Meteorological Studies, and School of Meteorology, University of Oklahoma, Norman, Oklahoma (Manuscript received 30 November 2005, in final form 20 June 2006) ABSTRACT Since the late 1960s, the West African Sudan–Sahel zone (10°–18°N) has experienced persistent and often severe drought, which is among the most undisputed and largest regional climate changes in the last half-century. Previous documentation of the drought generally has used monthly, seasonal, and annual rainfall totals and departures, in a standard “climate” approach that overlooks the underlying weather system variability. Most Sudan–Sahel rainfall occurs during June–September and is delivered by westwardpropagating, linear-type, mesoscale convective systems [disturbance lines (DLs)] that typically have much longer north–south (102–103 km) than east–west (10–102 km) dimensions. Here, a large set of daily rainfall data is analyzed to relate DL and regional climate variability on intraseasonal-to-multidecadal time scales for 1951–98. Rain gauge–based indices of DL frequency, size, and intensity are evaluated on a daily basis for four 440-km square “catchments” that extend across most of the West African Sudan–Sahel (18°W–4°E) and are then distilled into 1951–98 time series of 10-day and seasonal frequency/magnitude summary statistics. This approach is validated using Tropical Applications of Meteorology Using Satellite Data (TAMSAT) satellite IR cold cloud duration statistics for the same 1995–98 DLs. Results obtained for all four catchments are remarkably similar on each time scale. Long-term (1951–98) average DL size/organization increases monotonically from early June to late August and then decreases strongly during September. In contrast, average DL intensity maximizes 10–30 days earlier than DL size/ organization and is distributed more symmetrically within the rainy season for all catchments except the westernmost, where DL intensity tracks DL size/organization very closely. Intraseasonal and interannual DL variability is documented using sets of very deficient (8) and much more abundant (7) rainy seasons during 1951–98. The predominant mode of rainfall extremes involves near-season-long suppression or enhancement of the seasonal cycles of DL size/organization and intensity, especially during the late July– late August rainy season peak. Other extreme seasons result solely from peak season anomalies. On the multidecadal scale, the dramatic decline in seasonal rainfall totals from the early 1950s to the mid-1980s is shown to result from pronounced downtrends in DL size/organization and intensity. Surprisingly, this DL shrinking–fragmentation–weakening is not accompanied by increases in catchment rainless days (i.e., total DL absence). Like the seasonal rainfall totals, DL size/organization and intensity increase slightly after the mid-1980s.

1. Introduction The drought conditions that have afflicted the West African Sudan–Sahel zone (10°–18°N) since the late 1960s are recognized to be among the most undisputed * Current affiliation: International Research Institute for Climate and Society, The Earth Institute at Columbia University, Palisades, New York.

Corresponding author address: Prof. Peter J. Lamb, Cooperative Institute for Mesoscale Meteorological Studies, University of Oklahoma, 120 David L. Boren Blvd., Suite 2100, Norman, OK 73072-7304. E-mail: [email protected]

© 2006 American Meteorological Society

JCLI4020

and largest regional climate changes experienced on the earth during the last half-century (Dai et al. 2004). This situation is documented in Fig. 1, the principal features of which now are well known. They include extremely wet years through most of the 1950s, followed by a progressive rainfall decline that first produced drought conditions in 1968 and culminated in pulses of severe drought in 1971–73, 1977, 1982–84, and 1987. Since the late 1980s, there has been only a modest rainfall recovery, many very dry years, and adverse societal impacts reported as recently as mid-2005. That Fig. 1 depicts regional climate change rather than variability is established by the sequence of its subperiod-averaged departures—from ⫹0.92␴ (1950–58), to ⫹0.36␴ (1959–67),

5344

JOURNAL OF CLIMATE

VOLUME 19

FIG. 1. Time series (1941–2004) of average normalized April–October rainfall departure (␴) for 20 stations in the West African Sudan–Sahelian zone (11°–18°N) west of 10°E. Renormalized (for 1941–2000) and updated from earlier versions in Lamb (1978a,b, 1981, 1985), Lamb and Peppler (1991, 1992), and Tarhule and Lamb (2003), where further details can be found. This index has been shown to correlate strongly with other zonewide rainfall indices based on many more stations (Dai et al. 2004, ⫹0.95; Fall et al. 2006, ⫹0.91).

to ⫺0.51␴ (1968–87), to ⫺0.25␴ (1988–2004)—the differences between which (except for 1968–87 versus 1988–2004) are significant at the 5% level according to a “pooled variance” two-tailed t test (Wilks 1995, 122– 124). A range of empirical and modeling studies have linked temporal components of the rainfall variability in Fig. 1 to sea surface temperature (SST) anomaly patterns in the tropical Atlantic (interannual) and tropical Pacific (El Niño time scale) Oceans, and also globally where an interhemispheric contrast is involved (multidecadal). Those studies include Lamb (1978a,b), Folland et al. (1986), Palmer (1986), Lamb and Peppler (1992), Palmer et al. (1992), Ward (1998), Giannini et al. (2003), and Hoerling et al. (2006). The development and format of Fig. 1 result from application of a procedure that has become standard in observational studies of regional tropical and subtropical rainfall variability on interannual-to-multidecadal time scales. This approach yields a single normalized departure time series averaged over multiple stations, for which the basic input typically consists of seasonal or annual station totals. Such a distinctly “climate” perspective overlooks the weather system variability that produces the seasonal or annual rainfall departures featured in such time series. In contrast, this paper documents changes in the weather system characteristics that were most immediately responsible for the striking

temporal features of Fig. 1 summarized above. Thus, our focus is on the final links in the teleconnection chain between the aforementioned basin- and globalscale SST anomaly patterns and Sudan–Sahel seasonal rainfall totals—the individual weather systems that deliver rainfall locally. We consider most of the last half of the twentieth century (1951–98; cf. Fig. 1). The uniqueness of West African Sudan–Sahel weather systems has been recognized for 60 yr (Hamilton and Archbold 1945; Schove 1946; Eldridge 1957). An image of a classic example of these westwardmoving, mesoscale convective systems appears in Fig. 2. Known as Lignes de Grains (LDGs) in French or Disturbance Lines (DLs, used henceforth) in English, these systems typically have much longer north–south (102–103 km) than east–west (10–102 km) dimensions. These DLs, which tend to be embedded within easterly waves (EWs) and can include squall lines, develop within a West African tropospheric structure characterized by shallow low-level southwesterly monsoon flow overlain by deep easterlies, including the midtropospheric African easterly jet (AEJ) and the uppertropospheric Tropical easterly jet (TEJ). The DL moisture is derived from the southwest monsoon flow, and DL westward propagation is steered by the overlying easterlies, especially the AEJ (Hastenrath 1991, 239–243). Our use of this broad DL designation for Sudan–Sahel weather systems is supported strongly by

15 OCTOBER 2006

BELL AND LAMB

FIG. 2. Radar reflectivity image of a DL approaching Niamey, Niger, on 20 Aug 1989. Niamey (13.50°N, 2.13°E) is located at the center of the Niger catchment (N) in Fig. 3. Shading indicates regions of strongest radar reflectivity. From Hastenrath (1991, p. 241); earlier obtained by the present second author from the Agence pour la Sécurité de la Navigation Aérienne en Afrique et á Madagascar (ASECNA) Forecast Office at Niamey Airport, Niger.

Fink et al.’s (2006) review and typing of those systems. Their review and follow-up analyses, coupled with earlier results in Le Barbé and Lebel (1997) and Lebel et al. (1997), suggest that the rainfall fraction in the present study region (Fig. 3) produced by DLs decreases equatorward from more than 90% in the arid Sahel (15°–17°N) to around 70% in the considerably more humid northern Sudan (⬃11°N). Interest in the characteristics and dynamics of the DLs and their associated EWs has burgeoned since the mid-1970s. Initial stimuli were provided by the realization that some DLs/EWs spawned Atlantic hurricanes (Simpson et al. 1968, 1969; Carlson 1969a,b); by the seminal paper of Burpee (1972) that linked DLs/EWs to the horizontal and vertical wind shear associated with the AEJ; by the worldwide publicity given to the enormous societal disaster caused by the 1971–73 (e.g., Wade 1974; Englebert 1974) and 1982–84 (e.g., Walsh 1984; Kerr 1985) pulses of severe drought; and by analyses invited by the availability of data from the Global Atmospheric Research Program (GARP) Atlantic Tropical Experiment (GATE) in 1974 (e.g., Aspliden et al. 1976; Payne and McGarry 1977; Reed et al. 1977; Norquist et al. 1977). The GATE-based work focused in particular on the DL–EW relationship and

5345

suggested that DLs are enhanced by upward vertical motion and latent heat release just ahead of the trough axis, especially as the DLs approach the West African Atlantic coast from the east. Further DL understanding was obtained during the late 1980s and 1990s from analyses of satellite data of increased diversity and resolution, including as part of the International Satellite Cloud Climatology Project (ISCCP). This research provided information on the preferred areas of orographically forced DL genesis (e.g., Hodges and Thorncroft 1997), on the importance of monsoon moisture depth and possibly localized moisture sources for DL development (e.g., Rowell and Milford 1993), and in a very limited sense (Julys of 1983–85) on DL interannual variability (Desbois et al. 1988). A comprehensive rain gauge–based inquiry into DL temporal variability was performed in support of the 1990–93 Hydrology–Atmosphere Pilot Experiment in the Sahel (HAPEX-Sahel) for a central Sahel area extending across southern Niger and eastern Burkina Faso (e.g., Le Barbé and Lebel 1997; Lebel et al. 1997). This work spanned 1950–89 and attributed the progressive decrease in Sahel rainfall documented in Fig. 1 to a decline in the mean number of rainfall events, rather than decreases in other monsoon parameters such as season length or mean event rainfall totals. The same conclusion was reached in follow-up research for a larger central West African region for 1950–90 (Le Barbé et al. 2002). Most recently, the growing understanding of the DL–EW relationship was summarized in Fink and Reiner (2003). In this study, we use a large set of daily rainfall data for 1951–98 to document the intraseasonal-to-multidecadalscale variability of key DL characteristics (frequency, size, and intensity) for most of the West African Sudan–Sahel zone. An Eulerian-type approach is employed, in which DL characteristics are inferred for four large fixed “catchments” from daily rain gauge totals produced by the passage of each system across a catchment. These daily statistics, in turn, yield seasonal values of key monsoon parameters that are compared on an intercatchment basis and assessed within the context of their multidecadal time series. Because this study spans nearly half a century, data availability was not sufficient to employ a more Lagrangian approach using either radar reflectivity measurements (sporadic in time and space) or satellite imagery (only available since late 1977). However, rain gauge–based DL statistics are validated using satellite IR cold cloud duration (CCD) statistics produced for the same DLs for 1995– 98 by the Tropical Applications of Meteorology Using Satellite Data (TAMSAT) Group at the University of

5346

JOURNAL OF CLIMATE

VOLUME 19

FIG. 3. Location of West African Sudan–Sahel daily rainfall stations (open circles) and regional catchments (square boxes) used to generate DL-related rainfall statistics. Letters designate the catchments (S is Senegal, M is Mali, B is Burkina Faso, and N is Niger). Area shaded in inset map of Northern Hemisphere Africa locates those nations. See Fig. 4 for year-to-year variability in the number of stations reporting in each catchment during 1951–98.

Reading. Although this study ends in 1998 due to daily rainfall data availability, Fig. 1 and its discussion above suggest strongly that Sudan–Sahel drought has persisted into the new century on a seasonal time scale. Because of this situation and the nature of the daily rain gauge results presented below, it is likely that the DL characteristics identified here for the 1990s prevailed also in recent years.

2. Data and methods a. Daily rainfall data DL characteristics were inferred from a large daily rainfall dataset developed from several sources. The bulk of the rainfall data were from a set assembled originally by Dr. M. V. K. Sivakumar at the International Crops Research Institute for the Semi-Arid Tropics (ICRISAT-Sahel; Niamey, Niger), which included long-term records (from station inception until 1987) for approximately 530 stations in Senegal, Mali, Burkina Faso, and Niger. In addition, all daily station rainfall totals available for 1988–98 were received from the Directions de la Météorologie Nationale of Senegal, Mali, Burkina Faso, and Niger. Figure 3 locates the 337 stations used in the present study. Individual station records varied greatly in length and completeness, but, as explained below, all available daily data were used in the present analyses. Because our emphasis is on DL statistical characterization, rather than estimating areal rainfall accumulation, no attempt was made to account for spatial and temporal station variability. June–September daily station rainfall totals were used to generate time series of DL-related indices for

four square Sudan–Sahel catchments (Fig. 3), each 440 km by 440 km. These catchments were centered over central Senegal (14°39⬘N, 15°25⬘W; 99 possible stations), Bamako (Mali; 12°38⬘N, 8°02⬘W; 78 possible stations), Kindi (Burkina Faso; 12°26⬘N, 2°02⬘W; 116 possible stations), and Niamey (Niger; 13°30⬘N, 2°08⬘E; 44 possible stations). Essentially the same DL-related results were obtained for similarly centered circular catchments with 110- and 220-km radii. Time series (1951–98) of the June–September seasonal average number of daily reporting stations in each catchment are presented in Fig. 4. By far, the Burkina Faso catchment contained the largest number of stations for the longest period of time, especially after the mid-1960s. The Niger catchment generally contained the smallest number of stations, but with little interannual variability. The Senegal and Mali catchments experienced sharp declines in the number of reporting stations from 1973 to 1974 and a subsequent recovery to pre-1974 totals in 1987. For Mali, the 1974–86 rainfall reporting station minimum resulted from adverse impacts of the continuing severe drought conditions on volunteer observers (rural exodus, resignation, death, and inconsistent reporting stemming from lack of rain) and some weaknesses in the administrative management of the stations (K. Konare, former National Director of Meteorology for Mali, 2006, personal communication). After 1986, increased rainfall and improved station network management led to more consistent reporting from Mali’s rainfall stations. The same reasons contributed to the post-1973 reporting station minimum for Senegal (A. Ndiaye, former National Director of Meteorology for Senegal, 2006, per-

15 OCTOBER 2006

5347

BELL AND LAMB

FIG. 4. Time series (1951–98) of the June–September seasonal average number of rainfall stations reporting each day in each catchment.

sonal communication), from which the recovery began earlier and was more progressive (Fig. 4). However, with one possible exception noted below, the 1974–86 minima in reporting stations for the Senegal and Mali catchments do not appear to have introduced any significant biases in the results presented subsequently.

b. Analysis of daily rainfall data To characterize the DLs, several DL index time series were developed for 1951–98 for each catchment from the above daily rainfall totals. Each index was calculated on a daily basis from 1 June to 30 September (i.e., for 122 days) of each year. The Daily Disturbance Extent Index (DDEI) was simply the percentage of reporting stations in a catchment that received rainfall above a trace on a given day (Lamb et al. 1998). The DDEI thus represents the horizontal extent (and likely degree of organization) of DL rainfall within a catchment and includes the seasonal cycle. Next, time series of several DL magnitude/ frequency statistics were generated from the raw

1951–98 DDEI time series for each catchment: (i) number of days per season when DDEI ⱖ 70% (frequency of large/well-organized DLs); (ii) number of days per season when 0% ⬍ DDEI ⱕ 30% (frequency of small/ poorly organized DLs); (iii) numbers of days per season when DDEI ⫽ 0% (frequency of total DL absence); and (iv) average nonzero daily DDEI value per season (average DL size). Thus, each catchment time series (i)–(iv) possessed one value per season. A very similar approach and associated (DDEI %) thresholds for characterizing DL occurrence/size/organization were employed in earlier Sahelian (southern Niger, Mathon et al. 2002) and Sudanian (central Benin, Fink et al. 2006) studies, albeit for much smaller regions. Following Lamb et al. (1998), a Daily Disturbance Intensity Index (DDII) was defined for each catchment for each day j of each season k with rainfall above a trace as DDIIjk ⫽

1 Njk

Njk

兺 i⫽1

rijk ⫺ ri . ␴i

5348

JOURNAL OF CLIMATE

Here, rijk is the daily rainfall total (⬎trace) at station i on Julian day j in a particular season k, ri is the mean of all daily rainfall totals (⬎trace) recorded at station i during the June–September periods of 1951–98, ␴i is the standard deviation of all daily rainfall totals (⬎trace) recorded at station i during the June–September periods of 1951–98, and Njk is the number of available rainfall stations within the catchment that received rainfall (⬎trace) on Julian day j in season k. The DDII thus is not defined on those days when all available stations within a catchment did not receive rainfall above a trace, and so the index is a true measure of intensity. Like the DDEI, the DDII retains the seasonal cycle. Standardized departures were used here because of the strong south–north seasonal rainfall gradient in the region (Lamb 1978a). Consistent with the above DDEI analyses, time series of several DL magnitude/ frequency statistics were generated from the raw DDII time series for each catchment: (i) number of days per season when DDII ⱖ ⫹0.4␴ (frequency of strong DLs); (ii) number of days per season when DDII ⱕ ⫺0.4␴ (frequency of weak DLs); and (iii) average DDII value per season (average DL intensity). These catchment time series also contained one value per season. To facilitate investigation of interannual and multidecadal variability, the seasonal cycle was removed from the DDII to yield a Daily Disturbance Intensity Anomaly Index (DDIAI) that was defined (Lamb et al. 1998) for each catchment for each day j of each season k with rainfall above a trace as DDIAIjk ⫽

1 Njk

Njk

兺 i⫽1

rijk ⫺ 关ri兴 . 关␴i兴

Here, the terms are as defined for the DDII except for the square brackets that indicate the 1951–98 daily rainfall averages and standard deviations are for moving 5-day periods centered on Julian day j. The raw DDIAI time series yielded a counterpart set of DL frequency/ magnitude time series to those described above for the DDII. To permit the satellite validation described below, an additional DL intensity time series was generated. Catchment Average Rainfall (CAR) was calculated as the average of the daily rainfall totals from all reporting stations on a given day, with zero amounts included. Despite the lack of standardization and inclusion of zero values, intraseasonal CAR variations often resembled those of the DDII, and sometimes the DDEI as well. Likewise, the 1951–98 time series of seasonal average CAR displayed the same multidecadal decline and similar interannual variations as the time series of the seasonal average DDIAI, as will be shown.

VOLUME 19

Finally, to facilitate documentation of intraseasonal variability, all of the above time series of daily DL index values were reduced to time series of 10-/11-day averages. Each month during June–September was divided into three 10-day periods (except July and August, for which the third period included 11 days), and average rainfall indexes were calculated for each period that henceforth are termed “dekads.” The above Eulerian monitoring of migratory DLs using daily rainfall totals may have several limitations. Because the catchments are anchored in space and have a fixed size, not all of an individual DL might traverse a given catchment and, thus, the full DL size and/or degree of organization may not be reflected in the above statistics. However, the large catchments used were designed with cognizance of the typical DL orientation, path, and maximum north–south dimension (Fig. 2). Daily rainfall totals could reflect the passage of more than one DL through a catchment during a day. Alternatively, since many DLs are long-lived, rainfall from one DL could be spread over two consecutive days. However, the effect of this division could be reduced because daily rainfall measurements typically are made around 0600 UTC, which captures adequately the late night-to-early morning diurnal rainfall cycle maximum (e.g., Hastenrath 1991, p. 21 and p. 378; Shinoda et al. 1999). Despite these possible limitations, the DL results presented below are very similar across the four catchments.

c. Satellite data Since the mid-1980s, the TAMSAT group at the University of Reading (United Kingdom) has been developing and applying rainfall estimation techniques for parts of Africa based upon CCD data derived from thermal infrared images from Meteosat geostationary satellites (Dugdale 1994). This effort has supported operational, real-time drought monitoring for tropical North Africa, including the West African Sudan–Sahel zone, that dates back to 1988. The methods described below to validate the above gauge-based catchment rainfall statistics are derived from techniques developed by the TAMSAT group. This validation used daily CCD data for each catchment in Fig. 3. The daily CCD values were determined for each pixel in geostationary satellite infrared images—which measure 5 km ⫻ 5 km at the subsatellite point (0° latitude–longitude) and are slightly larger for our catchments—by summing the number of halfhourly images per day when the pixel contained a radiance value equivalent to a temperature equal to or below a chosen threshold. These daily CCD totals were compiled by the TAMSAT project to correspond with

15 OCTOBER 2006

5349

BELL AND LAMB

TABLE 1. Comparison of daily rain gauge–based DDEI and daily satellite-equivalent DDEI values, as functions of satellite threshold temperature (°C), for catchments in Fig. 3 for June–September periods of 1995–98. Columns under (a) give correlation coefficients, (b) contain RMSEs (as percent of rain gauge DDEI), and (c) present mean-square-error bias values (as percent of rain gauge DDEI). Maximum correlation values and minimum error and bias values for each season in each catchment are indicated in bold. (a)

(b)

Year

⫺40°C

⫺50°C

⫺60°C

⫺40°C

⫺50°C

1995 1996 1997 1998

⫹0.84 ⴐ0.84 ⫹0.79 ⫹0.83

ⴐ0.86 ⴐ0.84 ⫹0.81 ⴐ0.85

⫹0.82 ⫹0.77 ⴐ0.82 ⫹0.82

22.81 22.19 27.82 22.89

1995 1996 1997 1998

ⴐ0.87 ⴐ0.84 ⫹0.86 ⫹0.81

⫹0.85 ⴐ0.84 ⴐ0.89 ⴐ0.82

⫹0.78 ⫹0.77 ⫹0.87 ⫹0.80

20.91 22.16 25.34 25.08

1995 1996 1997 1998

⫹0.89 ⫹0.81 ⫹0.85 ⫹0.86

⫹0.89 ⴐ0.84 ⴐ0.86 ⴐ0.87

ⴐ0.90 ⫹0.82 ⫹0.83 ⫹0.84

Burkina 24.07 26.30 25.93 25.73

1995 1996 1997 1998

⫹0.81 ⫹0.85 ⫹0.77 ⴐ0.79

⫹0.84 ⴐ0.87 ⴐ0.79 ⫹0.78

ⴐ0.85 ⫹0.82 ⫹0.75 ⫹0.75

28.31 22.88 27.52 26.52

(c) ⫺60°C

⫺40°C

⫺50°C

⫺60°C

17.79 15.13 14.35 15.87

⫹13.31 ⫹13.45 ⫹18.07 ⫹13.04

ⴐ3.79 ⫹6.06 ⫹9.16 ⫹5.21

⫺8.34 ⴑ1.55 ⴑ0.47 ⴑ3.95

17.90 17.64 18.34 19.99

20.67 20.19 14.87 19.87

⫹11.74 ⫹13.11 ⫹19.45 ⫹15.81

ⴐ3.79 ⴐ3.96 ⫹10.19 ⴐ6.31

ⴑ8.91 ⴑ8.85 ⴑ2.16 ⫺6.41

Faso 17.55 19.44 18.84 19.49

13.11 16.53 14.66 16.72

⫹16.66 ⫹16.32 ⫹19.33 ⫹17.66

⫹8.05 ⫹8.05 ⫹9.98 ⫹8.99

ⴑ4.87 ⴑ3.03 ⴑ1.35 ⴑ2.35

15.61 14.93 13.66 18.80

⫹19.18 ⫹14.60 ⫹19.63 ⫹15.74

⫹10.32 ⫹6.05 ⫹10.22 ⫹6.88

ⴑ1.08 ⴑ3.93 ⴑ0.33 ⴑ3.20

Senegal 16.33 16.54 20.37 17.01 Mali

Niger

the typical daily observation period of West Africa rain gauges, namely 0600 to 0600 UTC. All daily CCD data available for June–September 1995–98 for the ⫺40°, ⫺50°, and ⫺60°C thresholds were provided by TAMSAT for this study. This range of temperature thresholds is the same as that used in the West African Sudan study of Fink et al. (2006), and numerous other investigations of deep convection/substantial rainfall referenced therein. TAMSAT used the software packages IDA [developed for the Famine Early Warning System (FEWS)] and WinDisp3 [developed for the Global Information and Early Warning System (GIEWS) Program at the Food and Agriculture Organization (FAO); http://www.fao.org/giews/english/ windisp/history.htm; http://www.fao.org/giews/english/ windisp/partners.htm] to generate the necessary CCD statistics from the raw radiance images. Despite the availability of just four concurrent seasons of daily rainfall and CCD data, Fig. 1 shows that these seasons span a sufficient range of Sudan–Sahelian rainfall—from the seventh driest since 1941 (1997) to the fifth wettest since the drought began in 1968 (1998).

d. Satellite validation of rain gauge–based indices To validate our DDEI for each catchment, a “satellite-equivalent” DDEI was calculated on a daily basis

21.47 15.99 19.04 21.53

simply by determining the percentage of total pixels in each catchment that had a CCD value greater than zero for a given temperature threshold. This validation procedure was based on the standard TAMSAT assumption that the surface beneath any pixel with a daily CCD value greater than zero for the chosen temperature threshold received some rainfall during the day (Dugdale 1994). The above satellite-equivalent DDEI thus gives an estimate of the proportion of the catchment that received rainfall on a given day. To test the validation for all three above temperature thresholds, the satellite-derived DDEI proxy was calculated on a daily basis using all available CCD data from June to September 1995–98, and compared with the daily rain gauge–based DDEI values for the same period. Table 1 presents statistics from those comparisons. For all above years, catchments, and temperature thresholds, there were very high correlations between rain gauge–based DDEI and satellite proxy DDEI values (Table 1a). Usually, the correlations were highest for the ⫺50°C threshold. Typically, for the warmer temperature thresholds there is a mean error bias (Wilks 1995, p. 254) toward higher satellite-equivalent DDEI values (Table 1c). CCD extent estimates based on warmer threshold temperatures cover a larger (or equal) proportion of a catchment than estimates based

5350

JOURNAL OF CLIMATE

VOLUME 19

TABLE 2. Example of contingency tables used to determine the most appropriate CCD threshold temperature for prediction of CAR. Values are numbers of days for the Niger catchment for Junes of 1996–98; selected threshold (⫺40°C) was used to predict CAR for June 1995. CCD temperature threshold ⫽ ⫺40°C Mean CCD ⫽ 0 Mean CCD ⬎ 0 Mean CAR ⫽ 0 7 3 Mean CAR ⬎ 0 9 69 CCD temperature threshold ⫽ ⫺50°C Mean CCD ⫽ 0 Mean CCD ⬎ 0 Mean CAR ⫽ 0 8 2 Mean CAR ⬎ 0 11 67 CCD temperature threshold ⫽ ⫺60°C Mean CCD ⫽ 0 Mean CCD ⬎ 0 Mean CAR ⫽ 0 9 1 Mean CAR ⬎ 0 16 62

FIG. 5. (a) Time series and (b) associated scatterplot of daily values of rain gauge–based DDEI (solid line, percent) vs satellitebased DDEI proxy (⫺50°C threshold; broken line, percent) for the Burkina Faso catchment for June–September 1998. Regression line has correlation coefficient of ⫹0.87 with statistical significance at 0.01% level (two-tailed t test) after accounting for autocorrelation and reducing effective sample size (Wilks 1995, 127–129).

upon a colder threshold temperature. Root-meansquare error (RMSE) and mean error (bias) calculations usually established that the smallest differences between the rain gauge–based DDEI and satelliteproxy DDEI estimates resulted from use of the coldest (⫺60°C) threshold (Tables 1b, c). The above strong correlations between the rain gauge–based DDEI and satellite-proxy DDEI values, and bias toward higher satellite-equivalent values, are illustrated well by the Burkina Faso catchment results for 1998 using the ⫺50°C threshold (Fig. 5). During the 120 days in 1998 for which both satellite and rain gauge data were available for Burkina Faso, the two measures of DL size tended to covary from day to day in a very similar way, despite several days differing substantially. As demonstrated by the correlation, RMSE, and bias values in Table 1, this example was typical of all catchments for the seasons considered. This situation, coupled with the fact that the differences between the gauge and satellite estimates of DL size were smaller for the lower temperature thresholds associated with

deeper convection, gives credence to our use of a rain gauge–based DDEI to quantify DL size. Validation of DL intensity was more challenging. The absence of long time series of infrared satellite data precluded direct validation of the DDII and DDIAI, since that required long-term averages and standard deviations to calculate daily values. Because CAR did not have that requirement, we used it to validate DL intensity employing the TAMSAT procedure (Diop 1998; Grimes and Diop 2003; Diop and Grimes 2003) for calculating the equivalent of CAR on a daily basis from satellite CCD data. Several steps were involved in the generation of a set of regression equations for each catchment and calendar month from data for three of the four seasons during 1995–98. Those equations then were used to make a “prediction” of CAR from CCD for the same calendar month in the remaining (independent) season. This monthly resolution balanced the needs of accommodating intraseasonal variability and having sufficient pairs of CAR/CCD data points to yield robust and statistically significant regression relationships. The first step involved determining CCD values for pixels containing catchment rainfall stations and averaging them on a daily basis for each catchment. These daily CCD values then were incorporated into contingency tables for each catchment–month–temperature threshold that related the number of days when CCD and CAR averages were in zero and nonzero categories. An example of these tables appears in Table 2, which contains the Niger catchment information for the Junes of 1996–98 from which the regression equation used to predict the June 1995 CAR was developed. The temperature threshold selected for each catchment– prediction month from the contingency tables maxi-

15 OCTOBER 2006

BELL AND LAMB

5351

TABLE 3. Summary of seasonal (June–September) verification statistics for linear regression prediction of CAR from satellite CCD data for catchments in Fig. 3 for 1995–98. Statistics for each season were calculated from the daily predictions made using separate equations for each month. (a) June–September correlation coefficient, (b) RMSEs (mm day⫺1), (c) bias statistics (mm day⫺1), and (d) the range of number of data pairs used to construct monthly regression relationships in each season. Year

(a)

(b)

(c)

(d)

1995 1996 1997 1998

⫹0.77 ⫹0.85 ⫹0.83 ⫹0.84

Senegal 3.43 2.06 2.43 2.87

⫺0.02 ⫹0.10 ⫹0.73 ⫺0.50

72–82 59–77 61–85 61–79

1995 1996 1997 1998

⫹0.82 ⫹0.84 ⫹0.73 ⫹0.81

Mali 3.36 3.21 4.28 3.60

⫺0.41 ⫺0.37 ⫹0.29 ⫹0.58

76–88 65–80 64–87 66–80

1995 1996 1997 1998

⫹0.90 ⫹0.76 ⫹0.80 ⫹0.88

Burkina Faso 2.85 3.51 2.64 2.45

⫺0.64 ⫺0.61 ⫹0.89 ⫹0.17

76–86 65–78 66–83 66–78

1995 1996 1997 1998

⫹0.83 ⫹0.79 ⫹0.57 ⫹0.71

Niger 2.87 2.82 3.11 3.99

⫺0.01 ⫺0.01 ⫹1.10 ⫺1.00

71–83 62–78 62–79 64–75

mized the simultaneous occurrence of nonzero daily average CCD and CAR and came closest to equalizing the frequencies of “cold cloud without rain” (CCD ⬎ 0, CAR ⫽ 0) and “rain without cold cloud”(CAR ⬎ 0, CCD ⫽ 0). For the example in Table 2, this resulting threshold was ⫺40°C. The next step was the generation of linear regression equations that yielded CAR for each catchment for each (independent) individual month from the corresponding (dependent) CCD values for the previously determined temperature threshold. This procedure excluded the cases when CCD ⬎ 0/CAR ⫽ 0 and CAR ⬎ 0/CCD ⫽ 0. For the Niger case documented in Table 2, 76 CAR/CCD data pairs (of a possible maximum 88; data were missing for 2 days) were available to construct the regression relation. The smallest/largest number of data pairs used in this procedure was 59/88 (Table 3). The correlation, root-mean-square error, and mean error (bias) characteristics of the resulting regression equations also are documented in Table 3. The correlations generally are quite high (⫹0.57 to ⫹0.90) and the mean errors small (⫺0.01 to ⫹1.10 mm). Figure 6 provides a representative scatterplot of the outcome of this regression procedure.

FIG. 6. Scatterplot of daily CAR predictions from CCD data vs rain gauge–based CAR (mm day⫺1) for Niamey catchment for June–September 1995. Daily predictions were made using separate linear regression equations for each month, as explained in text. The CCD temperature threshold used (⫺40°C) was the same for each month. Regression line has correlation coefficient of ⫹0.83 with statistical significance level of 0.001% (two-tailed t test); both CAR time series had essentially no autocorrelation (Wilks 1995, 127–129).

3. Results a. Climatological seasonal cycles of DL characteristics To provide a reference frame for the short time-scale (interannual, composite, and epoch) results that follow, Fig. 7 documents the long-term (1951–98) average seasonal cycles of the DDEI, DDII, and CAR defined above for the four Sudan–Sahel catchments in Fig. 3. The climatologies presented update those of Finch (1998), which covered only 1951–87. The DDEI seasonal cycles in Fig. 7 suggest that DL size/organization increases monotonically from early June through the last dekad in August, after which they decline much more rapidly during September. This pattern characterizes every catchment with the slight exception of Mali (Fig. 7b), for which the late August and early September DDEI values are essentially equal. Note that the September decline in DL size/organization produces late September DDEI values comparable to those of early June for all catchments except Senegal (Fig. 7a), signifying that this aspect of the monsoon retreat generally is 3 times more rapid than its advance. Among the four catchments, Mali is characterized by the highest maximum DDEI, Senegal experiences the largest seasonal DDEI range, and Niger (Fig. 7d) has the lowest maximum DDEI and smallest seasonal DDEI range. Figure 7 suggests that DL intensity maximizes sea-

5352

JOURNAL OF CLIMATE

VOLUME 19

FIG. 7. Long-term (1951–98) average June–September seasonal cycles of DDEI (solid line, %, left ordinate), DDII (broken line, ␴, right inside ordinate), and CAR (dotted line, mm day⫺1, right outside ordinate) for (a) Senegal, (b) Mali, (c) Burkina Faso, and (d) Niger catchments. Abscissa is labeled in month/dekad. All curves are reproduced in Figs. 8–11, where space permits the addition of 95% confidence intervals to each 1951–98 dekad-average DL index value. Boxes contain linear correlation coefficients between seasonal cycles of DL indices shown. Although the long-term dekadal averages contain information from 48 yr of data, the small number of dekads (12) and large autocorrelations in the seasonal cycles of the DL indices made it impossible (due to negative degrees of freedom) to determine correlation thresholds for given significance levels if the autocorrelations were taken into account. Without adjusting for autocorrelation and reducing effective sample size, correlation threshold for 5% (1%) statistical significance level is r ⱖ ⫹0.50 (⫹0.66) (Keeping 1995, p. 295; Wilks 1995, 127–129).

sonally before DL size/organization in the central part of the West African Sudan–Sahel. The Mali and Burkina Faso catchments experience DDII seasonal maxima in late July and early August, respectively, 2–3 dekads prior to the DLs becoming largest/most organized there (Figs. 7b,c). This separation is only 1 dekad for Niger (Fig. 7d), whereas the DDEI and DDII maxima coincide for the westernmost Senegal catchment (Fig. 7a). Thus, with the exception of Senegal, the DDII is distributed more symmetrically within the

rainy season than the DDEI. This difference is reflected in the correlations between the curves in Fig. 7, which are lowest (⫹0.84 to ⫹0.88) for the DDEI–DDII association for Mali, Burkina Faso, and Niger. The maximum seasonal DDII value is similar for all catchments except Senegal, where it is lower and associated with the smallest seasonal DDII range. The largest seasonal DDII range is characteristic of Niger, for which the seasonal maximum DDII is highest. The CAR seasonal cycles in Fig. 7 generally are con-

15 OCTOBER 2006

5353

BELL AND LAMB

TABLE 4. Selection of extreme seasons for analysis of intraseasonal and interannual DL variability, using daily rainfall indices (DDEI and DDIAI) and normalized seasonal index (Fig. 1). Dry seasons

Most extreme

Less extreme

Daily based ranking

Seasonally based ranking

1984 1983 1982 1972 1990 1973 1997 1987 1968

1983 1984 1977 1987 1972 1982 1997 1990 1986

Wet seasons Final choice

Daily based ranking

Seasonally based ranking

Final choice

1983 1984 1977 1987 1972 1982 1997 1990 1968

1952 1958 1964 1955 1954 1953 1961 1957 1967

1952 1955 1953 1954 1957 1958 1964 1951 1962

1952 1955 1953 1954 1957 1958 1964 1967 1994

sistent with those described already for the DDEI (especially) and DDII. This is characteristic of the Senegal catchment in particular, for which the CAR and DDEI curves are essentially identical (r ⫽ ⫹0.996) and feature early to mid-August plateaus and sharp peaks in late August (Fig. 7a). Senegal CAR also tracks closely the DDII there (r ⫽ ⫹0.97), but with a slightly earlier (late July–early August) plateau and less pronounced late August peak. For Mali, the CAR curve contains the extended August maximum characteristic of the DDEI and DDII, followed by a similarly precipitous September decrease (Fig. 7b). Accordingly, the three Mali DL seasonal cycles in Fig. 7 are highly correlated (⫹0.96, ⫹0.97). The CAR peak for Burkina Faso (mid-August) is slightly later than the DDII maximum there (early August), but still ahead of that catchment’s late August DDEI maximum (Fig. 7c). For Niger, the CAR and DDII maxima coincide in mid-August (Fig. 7d). Accordingly, the seasonal CAR cycles for the two easternmost catchments correlate strongly with their DDEI (⫹0.97, ⫹0.98) and DDII (⫹0.93, ⫹0.94) counterparts. CAR is lowest in early June in Senegal, because of that catchment’s northern location, and maximizes in Mali in early August.

b. Intraseasonal and interannual DL variability This section documents the intraseasonal and interannual variability of Sudan–Sahelian DLs using selections of very deficient and much more abundant rainy seasons from 1951–98. For each catchment, decadal DDEI and DDII averages for the selected years were compared with long-term (1951–98) mean values of the indices to identify key intraseasonal departures of DL size and intensity and provide insight into the resulting interannual variability. The selection of the study years involved several steps. First, seasonal DDEI and DDIAI (not DDII, to

emphasize interannual variability) means were calculated for each catchment for each season, and then averaged across all catchments for a given season because of the strong zonal coherence of extreme seasonal rainfall anomalies (e.g., Lamb 1981, 1985; Nicholson 1986; Nicholson and Palao 1993). Next, the resulting DDEI and DDIAI seasonal means were ranked separately across 1951–98, and the two rankings for each season then were averaged to yield a final ranking from the wettest (large/organized, intense DLs) to driest (small/ unorganized, weak DLs) season. As documented in Table 4, a daily DL index-based ranking was used in conjunction with the seasonal (April–October) standardized anomaly index in Fig. 1 to select the final sets of nine extremely wet and nine extremely dry seasons. Those final sets of extreme seasons included the eight driest (in order: 1983, 1984, 1977, 1987, 1972, 1982, 1997, and 1990) and seven wettest (1952, 1955, 1953, 1954, 1957, 1958, and 1964) seasons during 1951–98 in Fig. 1. Of these eight driest (seven wettest) seasonal index seasons, seven (six) were among the eight driest (seven wettest) seasons in the daily based ranking (Table 4). Thus, there was substantial agreement between the daily and seasonally based choices of extreme seasons for this intraseasonal and interannual inquiry. The two extremely dry seasons chosen that were not in both “top nine” rankings in Table 4 were 1968 and 1977. Selection of 1968 resulted from its ranking as the ninth-driest season using the daily index-based results, and being the focus of an earlier case study investigation with respect to forcing from tropical Atlantic SST anomalies (Lamb 1978a, 1983). Several interesting circumstances prompted selection of 1977. It was the third driest overall season in the Sudan–Sahel during 1941–98 (Fig. 1) and, as shown below, included a very large number of no-rain days in the two westernmost catchments. However, 1977 was not especially dry in

5354

JOURNAL OF CLIMATE

VOLUME 19

FIG. 8. Intraseasonal variation of selected dekad-average DL indices for (a) Senegal in 1958 (DDEI, solid line, percent), (b) Senegal in 1983 (DDII, solid line, ␴), (c) Mali in 1984 (DDII, solid line, ␴), and (d) Burkina Faso in 1994 (CAR, solid line, mm day⫺1). Broken line in each panel gives 1951–98 dekad-average values of DL index shown, for which vertical bars indicate 95% confidence intervals constructed using Student’s t distribution and standard deviations estimated from N ⫽ 48 values for each dekad (Keeping 1995, 183–184). Abscissa is labeled in month/dekad.

terms of the daily index-based rankings (Table 4). Finally, of the three remaining wet seasons ultimately chosen in Table 4, 1957 and 1967 were among the top nine wettest seasons according to the daily index-based rankings, and 1994 clearly stood out in the seasonal index (Fig. 1) as the overall wettest season in the Sahel since the end of the 1960s. Examination of the intraseasonal and interannual variability of DL size and intensity for each catchment in the above extreme dry and wet seasons revealed

common characteristics, as documented in Fig. 8. The mode of substantial dekad-to-dekad DDEI, DDII, and CAR variability is illustrated in Figs. 8a,b for Senegal for 1958 (extremely wet) and 1983 (extremely dry). Most prominent are the respective late July–late August extremes that largely determined the overall monsoon season quality, but which are flanked by dekads with opposite anomalies. Conversely, and more importantly, other extreme seasons were characterized by strong persistence of dekad anomalies of the same sign.

15 OCTOBER 2006

5355

BELL AND LAMB

TABLE 5. Intraseasonal variability of DL indices during extreme seasons identified in Table 4. Columns for each catchment give number of dekads (out of 12) during each extreme season (Table 4) when the dekad average DDEI, DDII, or CAR was above the 1951–98 average (extreme wet years, upper half of table) and below the 1951–98 average (extreme dry years, lower half). Cases in which fewer than half of the 12 dekads were above normal (extreme wet years) or below normal (extreme dry years) are in bold. Extreme right columns give all-catchment totals (out of 48) as percentages. Senegal

Mali

Burkina Faso

Niger

All catchments

DDEI DDII CAR DDEI DDII CAR DDEI DDII CAR DDEI DDII CAR DDEI DDII CAR Extreme 1952 wet years 1955 1953 1954 1957 1958 1964 1967 1994

8 11 7 8 6 7 7 10 4

8 9 7 8 7 7 8 6 5

9 9 8 7 7 7 8 8 4

7 11 7 9 8 9 10 8 8

9 6 5 9 8 8 10 10 8

9 9 7 10 10 9 9 11 10

11 8 9 8 8 9 10 8 7

8 6 8 8 7 8 8 5 9

9 7 8 8 6 9 8 7 9

10 9 8 6 9 7 9 8 8

11 10 8 7 5 8 9 10 9

11 7 9 7 7 7 10 9 10

75 81 65 65 65 67 75 71 56

75 65 58 67 56 65 73 65 65

79 67 65 67 63 67 73 73 69

Extreme 1983 dry years 1984 1977 1987 1972 1982 1997 1990 1968

9 7 5 7 9 9 9 9 9

9 8 7 6 10 9 6 10 10

9 7 7 8 10 9 9 10 10

8 9 8 11 9 10 8 7 6

9 11 9 7 11 6 5 7 8

11 11 8 10 10 9 5 6 9

10 11 7 9 7 11 11 11 4

7 11 3 6 10 10 6 6 9

10 11 8 8 8 11 9 11 6

10 10 7 11 9 11 11 11 7

8 9 6 9 8 8 8 8 7

8 10 7 11 7 10 9 9 9

77 77 56 79 71 85 81 79 54

67 81 52 58 81 69 52 65 71

79 81 63 77 73 81 67 75 71

For example, during the extremely dry 1984 season the DDII was below average for 11 of 12 dekads in Mali (Fig. 8c), whereas in the much wetter 1994 the Burkina Faso CAR was above average in 9 of 12 dekads (Fig. 8d). More generally, Table 5 shows a strong tendency for at least 65% of the catchment dekads during each extreme wet (dry) season to have above (below) average DL index values. A particularly remarkable result in Table 5 is that, in almost one-quarter (51 out of 216) of its catchment seasons, at least 10 of the 12 dekads had DL index anomalies of the expected sign. The most notable exception is the Senegal catchment for 1994, despite that season being the wettest for the West African Sudan–Sahel zone as a whole since 1969 (Fig. 1). Also evident in Table 5 is further evidence of the aforementioned unusual nature of the 1977 drought season. The intraseasonal and interannual variability of DL characteristics was investigated further by compositing on a catchment basis the dekad average DDEI, DDII, and CAR values for the two sets of extreme seasons in Tables 4 and 5. Results appear in Figs. 9–11. Across all catchments for most of the June–September season, the wet composite DDEI, DDII, and CAR values are markedly higher than the 1951–98 averages, and their dry composite counterparts are similarly lower. These differences are especially large from late July through late August, with most exceptions occurring during monsoon onset in June. Seasonal cycle composite

maxima generally coincide with their 1951–98 counterparts, or else occur in a contiguous dekad. However, while the wet composite seasonal cycles tend to be amplifications of the long-term average curves, the “arrested development” of the dry composite rainy season is indicated clearly by the flattening of many catchment DL index profiles across the normal rainy season peak. An additional investigation into intraseasonal and interannual DL variability was performed by dividing 1951–98 into continuous multiyear wet (1951–67) and dry (1968–98) epochs, separated by the onset of nearly continuous Sudan–Sahel drought as revealed by seasonal rainfall totals (Fig. 1). Analysis of the DL behavior characteristic of these contrasting epochs followed that employed in the above composite analysis of discrete extreme years. The resulting seasonal cycle profiles (not shown) broadly replicated their counterparts in Figs. 9–11, but with wet-versus-dry differences being less pronounced because each epoch included nonextreme years.

c. Multidecadal DL variability This section analyzes time series of the DDEI, DDIAI (not DDII, again to emphasize interannual variability), and CAR extreme magnitude/frequency statistics defined in section 2b, along with seasonal average values of those indices, to document the multidecadal variability of key aspects of DL behavior dur-

5356

JOURNAL OF CLIMATE

VOLUME 19

FIG. 9. Intraseasonal variation of dekad-average DDEI (%) for 1951–98 (solid line), extreme wet year composite (dotted line), and extreme dry year composite (broken line) for (a) Senegal, (b) Mali, (c) Burkina Faso, and (d) Niger catchments. Abscissa is labeled in month/dekad. Vertical bars straddling solid lines indicate 95% confidence intervals for 1951–98 dekad means, constructed using Student’s t distribution and standard deviations estimated from N ⫽ 48 values for each dekad (Keeping 1995, 183–184). Vertical bars above (below) wet (dry) composite lines give standard deviations of dekad-average values contributing to composite means.

ing 1951–98. Each time series used contained one value per season. Figure 12 documents the temporal variation of extremes of DL size/organization during 1951–98. Despite considerable interannual variability, there was a strong tendency for the frequency of large and well-organized (small or unorganized) DLs to decline (increase) substantially during most of 1951–98. These trends were especially striking for Senegal and Burkina Faso. For instance, the number of days per season when the Senegal catchment (Fig. 12a) experienced large DLs decreased from a maximum of 29 in the very wet 1955 (Fig. 1; Tables 4 and 5) to a minimum of only one in 1991, whereas its frequency of small/unorganized DLs increased from 6 days in (again) 1955 to 34 in the very dry 1984 (Fig. 1; Tables 4 and 5). While the decline in

large DL frequency occurred gradually over 30–40 yr in Senegal (and Niger), for Burkina Faso it largely was confined to the 1970s and separated epochs that were quite uniformly very wet and very dry (Fig. 12c). Also prominent in Fig. 12 are the highly contrasting DL frequencies for Niger for the very dry 1980–84 (Fig. 1; Tables 4 and 5). Large/well-organized DLs failed to occur in three out of these five seasons, and the other two seasons included only one and three days characterized by such DLs (Fig. 12d). In contrast, those seasons included maximum or near-maximum numbers of days (65–80) when Niger DLs were small/unorganized. For Mali and Niger (Figs. 12b,c), slight upward (downward) tendencies in the frequency of large (small) DLs are apparent after the mid-1980s, which may constitute the start of a reversal of the trends of the previous 35 yr.

15 OCTOBER 2006

BELL AND LAMB

5357

FIG. 10. Same as in Fig. 9, except for DDII (␴).

Surprisingly, the above pronounced decline (increase) in the frequency of large (small) DLs across all catchments for most of 1951–98 was not accompanied by any consistent zonewide increase in the number of rainless days for entire catchments (i.e., total DL absence; Fig. 13). Only for Niger is there a suggestion of any long-term increase in rainless days, with the post1958 trend there being rather slight (Fig. 13d). More pronounced increases characterized shorter periods for Senegal (1964–80, post-1989) and Mali (1972–86), with no trend apparent for Burkina Faso (Figs. 13a–c). However, the Senegal and Mali increases for 1974–86 may be questionable because of the aforementioned decreased number of rain gauges reporting in that period (Fig. 4). A particularly interesting feature of Fig. 13 is the spikelike maxima of rainless days for 1977 for the two westernmost catchments of Senegal and Mali.

These maxima appear to be real, since they help explain why 1977 was so dry on a seasonal zonewide basis (Fig. 1), despite its DLs not being especially unproductive when they actually occurred (Tables 4 and 5; Fig. 12, Figs. 14–16 below). A rainless day maximum contributed similarly to the 1980 dryness in Senegal (cf. Figs. 12a and 13a). Figure 14 shows that the above pronounced decrease in DL size/organization throughout most of 1951–98 was associated with a reduction in DL intensity. For Senegal, there are clear downward (upward) trends in the frequency of strong (weak) DLs spanning the entire period (Fig. 14a). However, the other catchments to the east show some evidence of a reversal of these (previously strong) trends since the mid-1980s (Figs. 14b–d). Many of the aforementioned extreme wet and dry seasons (Tables 4 and 5) and extended periods of severe

5358

JOURNAL OF CLIMATE

VOLUME 19

FIG. 11. Same as in Fig. 9, except for CAR (mm day⫺1).

drought (Fig. 1) were associated with pronounced extremes in the frequency of strong or weak DLs in most catchments. For instance, during the extreme drought years of the early 1970s and early to mid-1980s, the Mali, Burkina Faso, and Niger catchments experienced extreme minima (maxima) in the frequency of strong (weak) DLs. This was especially characteristic of the very dry 1972 and 1984 seasons. Conversely, the seasonal frequency of strong DLs maximized in the very wet 1952 for all three of the above catchments. A more comprehensive multidecadal perspective appears in Fig. 15, which presents 1951–98 time series of seasonal average DDEI, DDIAI, and CAR values for each catchment. These time series confirm further the above pronounced downtrends in DL size/organization and intensity from 1951 until at least the mid-1980s in all catchments. Clearly, it is these changes in DL behavior that produced the dramatic decline in seasonal

rainfall totals over the same period (Fig. 1). After the mid-1980s, the preceding downtrend in average DL intensity is reversed modestly for Mali, Burkina Faso, and Niger, as are the trends in average DL size for Mali and Niger (Figs. 15b–d). For Burkina Faso and Senegal, however, the downtrend in average DL size continued through 1998 (Figs. 15a,c). In Senegal, a similar longterm CAR decrease occurred, whereas DDIAI increased somewhat for that catchment after the mid1980s. The correlations in Fig. 15 quantify the interannual covariance between DL size and intensity that underpins their above shared multidecadal trends. These positive associations are strongest for the more northerly located Senegal and Niger catchments. The especially high DDEI–CAR correlations presumably reflect the retention of seasonal cycles in both indices and the inclusion of zero amounts in the CAR calculations (sec-

15 OCTOBER 2006

BELL AND LAMB

5359

FIG. 12. Time series (1951–98) of June–September seasonal frequency of large/well-organized DLs (number of days per season DDEI ⱖ 70%, solid line, left ordinate) and small/poorly organized DLs (number of days per season 0% ⬍ DDEI ⱕ 30%, broken line, right ordinate) for (a) Senegal, (b) Mali, (c) Burkina Faso, and (d) Niger catchments.

tion 2c). Further documentation of this DL size– intensity relationship appears in Fig. 16 in the form of catchment DDEI–DDIAI scatterplots in which each pair of seasonal average values is located and identified. In these plots, the preferred lower-left locations of several of Fig. 1’s driest overall seasons (1972, 1973, 1983, 1984, 1990, and 1997) reinforce further that their DLs were both very small/unorganized and rather weak. In contrast, the more central position of 1977 in Fig. 16 for all catchments except Senegal’s DL intensity offers additional evidence that, when DLs occurred in this season, they were of average size and intensity. As noted earlier, this situation was offset in Senegal and Niger by high frequencies of DL absence. At the other extreme, most of the very wet seasons (Fig. 1) in the 1950s (e.g., 1952, 1954, 1955, and 1958) and 1960s (e.g., 1964 and 1967) consistently appear in the upper right of the panels in Fig. 16, confirming further that their DLs were large, organized, and intense. As noted above,

1994 was wetter than the preceding 25 yr for the zone as a whole (Fig. 1). Figure 16 shows that the DLs in this year were well developed, except for their smallish size in Senegal.

4. Summary and discussion The drought conditions that have burdened the West African Sudan–Sahel zone (10°–18°N) since the late 1960s, with varying intensity, have attracted scientific attention for several reasons. Not only is this drought among the earth’s most undisputed and largest regional climate changes in the last half-century, but it has been linked very strongly to basin- and global-scale sea surface temperature (SST) anomaly patterns on interannual-to-multidecadal time scales. This paper has focused on the final links in the teleconnection chain between those SST anomaly patterns and Sudan–Sahel seasonal rainfall totals—the individual weather systems

5360

JOURNAL OF CLIMATE

VOLUME 19

FIG. 13. Time series (1951–98) of June–September seasonal frequency of DL absence (number of days per season DDEI ⫽ 0%). Note that rainless days occur more often across the Senegal and Niger catchments because of their more northerly location.

that deliver rainfall locally. Previous documentation of the drought generally used rainfall accumulations and departures (monthly, seasonal, and annual) in a standard “climate” approach that overlooked the underlying weather system variability. Here, we used a large set of daily rainfall data for 1951–98 to document the intraseasonal-to-multidecadalscale variability of key characteristics (frequency, size, and intensity) of West African mesoscale convective disturbance lines (DLs) for most of the Sudan–Sahel zone. This first comprehensive zonewide treatment of weather system variability involved four large (440 km square) “catchments” extending from the Atlantic coast of Senegal (18°W) eastward to Nigeria (4°E). This rain gauge–based approach was validated using satellite IR cold cloud duration (CCD) statistics produced for the same DLs for 1995–98 by the TAMSAT Group at the University of Reading. Results obtained for all four catchments were remarkably similar on each time scale. The seasonal cycle

of long-term (1951–98) average DL size/organization featured a monotonic increase from early June through late August, followed by a strong decrease during September. In contrast, average DL intensity maximized 10–30 days earlier than DL size/organization and was found to be distributed more symmetrically within the June–September rainy season for all catchments except the westernmost, where DL intensity tracked DL size/ organization very closely. We found little evidence of any pronounced “jump” in the seasonal cycles of the DL indices for all catchments, as has been noted for the monsoon’s northward advance from farther south in West Africa at the end of June (e.g., Le Barbé et al. 2002). Intraseasonal and interannual DL variability were documented using composite analysis for sets of very deficient (8) and more abundant (7) rainy seasons during 1951–98. The predominant mode of rainfall extremes identified involved near season-long suppression or enhancement of the seasonal cycles of DL size/

15 OCTOBER 2006

BELL AND LAMB

5361

FIG. 14. Time series (1951–98) of June–September seasonal frequency of strong DLs (number of days per season when DDIAI ⱖ ⫹0.4␴, solid line, left ordinate) and weak DLs (number of days per season when DDIAI ⱕ ⫺0.4␴, broken line, right ordinate) for (a) Senegal, (b) Mali, (c) Burkina Faso, and (d) Niger catchments.

organization and intensity, especially during the late July–late August rainy season peak. Other extreme seasons resulted solely from peak season anomalies. These results provide insight into earlier findings regarding the societal importance of peak season rainfall reduction that were based solely on monthly rainfall totals (e.g., Dennett et al. 1985; Lamb 1985). On the multidecadal scale, the dramatic decline in seasonal rainfall totals from the early 1950s through the mid-1980s was shown to result from pronounced downtrends in both DL size/organization and intensity. Surprisingly, this DL shrinking/fragmentation/weakening was not accompanied by increases in catchment rainless days (i.e., total DL absence). These results are consistent with Shinoda et al.’s (1999, p. 93) finding concerning midnight-to-morning rainfall events at Niamey— that “the amounts and frequencies increased for the wet periods (the 1950s and late 1980s to early 1990s) and decreased over the drought periods (early 1970s

and 1980s).” On the other hand, the present results appear to be in striking contrast to the earlier HAPEXSahel-related attribution of the above progressive Sahel rainfall decrease to a decline in the mean number of rainfall events, rather than to reductions in mean event rainfall totals or monsoon season shortening (e.g., Le Barbé and Lebel 1997; Lebel et al. 1997; Le Barbé et al. 2002). However, the explanation for this apparent difference may in part be matters of spatial scale and sampling. Our finding of progressive DL shrinking and/or fragmentation between the early 1950s and mid-1980s was obtained for large (440 km ⫻ 440 km) catchments through the use of all available rain gauges on a collective basis. This DL shrinking/fragmentation likely caused the individual rain gauges on which the earlier findings were based to have been traversed by a declining number of DLs (or DL components) than occurred previously. The “mean number” of recorded rainfall

5362

JOURNAL OF CLIMATE

VOLUME 19

FIG. 15. Time series (1951–98) of June–September seasonal average of DDEI (solid line, %, left ordinate), DDIAI (broken line, ␴, right inside ordinate), and CAR (dotted line, mm day⫺1, right outside ordinate) for (a) Senegal, (b) Mali, (c) Burkina Faso, and (d) Niger catchments. Boxes contain linear correlation coefficients between seasonal cycles of DL indices shown; correlation threshold for 5% (1%) statistical significance level is r ⱖ ⫹0.62 (⫹0.79) after accounting for autocorrelation and reducing effective sample size (Keeping 1995, p. 295; Wilks 1995, 127–129).

events would have decreased. Concerning DL intensity, our finding of a 30⫹ year decline was based on the frequency of extremes, whereas the earlier conclusion of little such change was for “mean event rainfall.” Finally, like the seasonal rainfall totals, DL size/ organization and intensity increased slightly from the mid-1980s through the late 1990s, but without approaching the levels of the early 1950s. The rain gauge–based results presented here suggest that pronounced changes in DL behavior and functioning occur or have occurred on intraseasonal-tointerannual time scales. These results should help focus investigations of the underlying mesoscale dynamics and thermodynamics. Of particular interest are the fi-

nal links in the teleconnection chain between the aforementioned basin- and global-scale SST anomaly patterns and DL development. Specifically, what were the immediate mesoscale causes of the above progressive multidecadal decrease of both DL size/organization and intensity, and how were those causes related to the now-accepted much-larger-scale forcing? The positive relation between DL size/organization and intensity is consistent with the theory for strong long-lived squall lines proposed by Rotunno et al. (1988), observational evidence presented by Fritsch et al. (1986) for mesoscale convective weather systems in the central United States, and recent three-dimensional simulations of those systems by Coniglio et al. (2006).

15 OCTOBER 2006

BELL AND LAMB

5363

FIG. 16. Scatterplots (1951–98) of June–September seasonal average DDIAI (␴) as a function of DDEI (%) for (a) Senegal, (b) Mali, (c) Burkina Faso, and (d) Niger catchments. Final two digits of years serve as markers for individual seasons.

Acknowledgments. This research benefited greatly from the assistance of several organizations and many individuals. Particularly important were the first author’s extended visits to the African Centre of Meteorological Applications for Development (ACMAD; Niamey, Niger) and the Tropical Applications of Meteorology Using Satellite Data Group (TAMSAT, University of Reading, United Kingdom). It is a pleasure to acknowledge the critical support provided during these visits by the former ACMAD Director (Mohammed Sadek Boulahya) and the TAMSAT Group Leader (Dr. David I. F. Grimes). The visits were funded by the International Activities Office of the U.S. National Weather Service, the then Director of which (Dr. Martin C. Yerg Jr.) provided crucial support. Dr. M. V. K. Sivakumar generously contributed a large set of daily rainfall data that he had constructed with great care while at ICRISAT-Sahel (Niamey). Other daily rainfall data were provided by the Directions de la Météorologie Nationale (DMN) of Senegal, Mali, Burkina Faso, and Niger. Kaliba Konare (former Director, Mali DMN) and Alioune Ndiaye (former Director, Senegal

DMN) furnished understanding of the rainfall recording practices for their countries. The satellite validation of the rain gauge–based results stemmed from a suggestion by Professor Keith A. Browning (University of Reading). We acknowledge also the varied assistance and guidance of colleagues at TAMSAT (Drs. Virginia Thorne and Mariane Diop), ACMAD (Omar Baddour, Stéphane Jamoneau, Issa Lele Mouhamadou, Didier Ouedrago, and Professor J. Bayo Omotosho), CIMMS (Professor Michael B. Richman, Dr. M. Neil Ward, Diane H. Portis, Jonathan D. Finch, and Dr. Pauline A. Dibi Kangah), NOAA’s National Severe Storms Laboratory (Drs. David J. Stensrud and Michael C. Coniglio), and Florida State University (Professor Sharon E. Nicholson). Constructive suggestions and comments by two formal reviewers—including Professor Kerry H. Cook, who identified herself—led to important strengthening of the manuscript. The production of the manuscript and final figures at CIMMS by Luwanda Byrd and Issa Lele Mouhamadou, respectively, are greatly appreciated. Additional funding was provided by NOAA Grant NA17RJ1227.

5364

JOURNAL OF CLIMATE REFERENCES

Aspliden, C. I., Y. Tourre, and J. B. Sabine, 1976: Some climatological aspects of West African disturbance lines during GATE. Mon. Wea. Rev., 104, 1029–1035. Burpee, R. W., 1972: The origin and structure of easterly waves in the lower troposphere of North Africa. J. Atmos. Sci., 29, 77–90. Carlson, T. N., 1969a: Synoptic histories of three African disturbances that developed into Atlantic hurricanes. Mon. Wea. Rev., 97, 256–276. ——, 1969b: Some remarks on African disturbances and their progress over the tropical Atlantic. Mon. Wea. Rev., 97, 716– 726. Coniglio, M. C., D. J. Stensrud, and L. J. Wicker, 2006: Effects of upper-level shear on the structure and maintenance of strong quasi-linear mesoscale convective systems. J. Atmos. Sci., 63, 1231–1252. Dai, A., P. J. Lamb, K. E. Trenberth, M. Hulme, P. D. Jones, and P. Xie, 2004: The recent Sahel drought is real. Int. J. Climatol., 24, 1323–1331. Dennett, M. D., J. Elston, and J. R. Rogers, 1985: A reappraisal of rainfall trends in the Sahel. J. Climatol., 5, 353–361. Desbois, M., T. Kayiranga, B. Gnamien, S. Guessous, and L. Picon, 1988: Characterization of some elements of the Sahelian climate and their interannual variations for July 1983, 1984, and 1985 from the analysis of METEOSAT ISCCP data. J. Climate, 1, 867–904. Diop, M., 1998: The influence of weather systems on satellite rainfall estimation with applications to riverflow modelling. Ph.D. dissertation, University of Reading, 225 pp. ——, and D. I. F. Grimes, 2003: Satellite-based rainfall estimation for river flow forecasting in Africa. II. African easterly waves, convection, and rainfall. Hydrol. Sci., 48, 585–599. Dugdale, G., 1994: Satellite derived rainfall estimates over Africa: Their validation and use in monitoring climate change. Global Precipitations and Climate Change, M. Desbois and F. Desalmand, Eds., NATO ASI Series, Vol. I 26, SpringerVerlag, 279–292. Eldridge, R. H., 1957: A synoptic study of West African disturbance lines. Quart. J. Roy. Meteor. Soc., 83, 303–314. Englebert, V., 1974: Drought threatens the Tuareg World. Natl. Geogr. Mag., 145, 544–571. Fall, S., F. M. H. Semazzi, D. D. S. Niyogi, R. O. Anyah, and J. Bowden, 2006: Spatiotemporal climate variability over Senegal and its relationship to global climate. Int. J. Climatol., 26, in press. Finch, J. D., 1998: Variability of tropospheric airflows over Subsaharan West Africa. M.S. thesis, School of Meteorology, University of Oklahoma, 105 pp. Fink, A. H., and A. Reiner, 2003: Spatiotemporal variability of the relation between African easterly waves and West African squall lines in 1998 and 1999. J. Geophys. Res., 108, 4332, doi:10.1029/2002JD002816. ——, D. G. Vincent, and V. Ermert, 2006: Rainfall types in the West African Sudanian Zone during the summer monsoon 2002. Mon. Wea. Rev., 134, 2143–2164. Folland, C. K., T. N. Palmer, and D. E. Parker, 1986: Sahel rainfall and worldwide sea temperatures. Nature, 320, 602–607. Fritsch, J. M., R. J. Kane, and C. R. Chelius, 1986: The contribution of mesoscale convective weather systems to the warmseason precipitation in the United States. J. Climate Appl. Meteor., 25, 1333–1345.

VOLUME 19

Giannini, A., R. Saravanan, and P. Chang, 2003: Oceanic forcing of Sahel rainfall on interannual to interdecadal time scales. Science, 302, 1027–1030. Grimes, D. I. F., and M. Diop, 2003: Satellite-based rainfall estimation for river flow forecasting in Africa. I. Rainfall estimates and hydrological forecasts. Hydrol. Sci., 48, 567–584. Hamilton, R. A., and J. W. Archbold, 1945: Meteorology of Nigeria and adjacent territory. Quart. J. Roy. Meteor. Soc., 71, 231–262. Hastenrath, S., 1991: Climate Dynamics of the Tropics. Kluwer, 488 pp. Hodges, K. I., and C. D. Thorncroft, 1997: Distribution and statistics of African mesoscale convective weather systems based upon the ISCCP Meteosat imagery. Mon. Wea. Rev., 125, 2821–2837. Hoerling, M. P., J. W. Hurrell, and J. Eischeid, 2006: Detection and attribution of twentieth-century northern and southern African monsoon change. J. Climate, 19, 3989–4008. Keeping, E. S., 1995: Introduction to Statistical Inference. Dover Publications, 451 pp. Kerr, R. A., 1985: Fifteen years of African drought. Science, 227, 1453–1454. Lamb, P. J., 1978a: Case studies of tropical Atlantic surface circulation patterns during recent sub-Saharan weather anomalies: 1967 and 1968. Mon. Wea. Rev., 106, 482–491. ——, 1978b: Large-scale tropical Atlantic surface circulation patterns associated with Sub-Saharan weather anomalies. Tellus, 30, 240–251. ——, 1981: Persistence of Subsaharan drought. Nature, 299, 46–48. ——, 1983: West African water vapor variations between recent contrasting Subsaharan rainy seasons. Tellus, 35A, 198–212. ——, 1985: Rainfall in Subsaharan West Africa during 1941–83. Zeit. Gletscherkunde Glacialgeol., 21, 131–139. ——, and R. A. Peppler, 1991: West Africa. Teleconnections Linking Worldwide Climate Anomalies: Scientific Basis and Societal Impacts, M. H. Glantz, R. W. Katz, and N. Nicholls, Eds., Cambridge University Press, 121–189. ——, and ——, 1992: Further case studies of tropical Atlantic surface atmospheric and oceanic patterns associated with sub-Saharan drought. J. Climate, 5, 476–488. ——, M. A. Bell, and J. D. Finch, 1998: Varibility of Sahelian disturbance lines and rainfall during 1951–87. Water Resources Variability in Africa during XXth Century, E. Servat et al., Eds., Publication 252, IAHS, 19–26. Le Barbé, L., and T. Lebel, 1997: Rainfall climatology of the HAPEX-Sahel region during the years 1950–1990. J. Hydrol., 188–189, 43–73. ——, ——, and D. Tapsoba, 2002: Rainfall variability in West Africa during years 1950–90. J. Climate, 15, 187–202. 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., H. Laurent, and T. Lebel, 2002: Mesoscale convective system rainfall in the Sahel. J. Appl. Meteor., 41, 1081–1092. Nicholson, S. E., 1986: The spatial coherence of African rainfall anomalies: Interhemispheric teleconnections. J. Climate Appl. Meteor., 25, 1365–1381. ——, and I. M. Palao, 1993: A re-evaluation of rainfall variability in the Sahel. Part I. Characteristics of rainfall fluctuations. Int. J. Climatol., 13, 371–389. Norquist, D. C., E. E. Recker, and R. J. Reed, 1977: The energetics of African wave disturbances as observed during Phase III of GATE. Mon. Wea. Rev., 105, 334–342.

15 OCTOBER 2006

BELL AND LAMB

Palmer, T. N., 1986: The influence of the Atlantic, Pacific, and Indian Oceans on Sahel rainfall. Nature, 322, 251–253. ——, C. Brankovic´ , P. Viterbo, and M. J. Miller, 1992: Modeling interannual variations of summer monsoons. J. Climate, 5, 399–417. Payne, S. W., and M. M. McGarry, 1977: The relationship of satellite inferred convective activity to easterly waves over West Africa and the adjacent ocean during Phase III of GATE. Mon. Wea. Rev., 105, 413–420. Reed, R. J., D. C. Norquist, and E. E. Recker, 1977: The structure and properties of African wave disturbances as observed during Phase III of GATE. Mon. Wea. Rev., 105, 317–333. Rotunno, R., J. B. Klemp, and M. L. Weisman, 1988: A theory for strong, long-lived squall lines. J. Atmos. Sci., 45, 463–485. Rowell, D. P., and J. R. Milford, 1993: On the generation of African squall lines. J. Climate, 6, 1181–1193. Schove, D. J., 1946: A further contribution to the meteorology of Nigeria. Quart. J. Roy. Meteor. Soc., 72, 105–110. Shinoda, M., T. Okatani, and M. Saloum, 1999: Diurnal variations

5365

of rainfall over Niger in the West African Sahel: A comparison between wet and drought years. Int. J. Climatol., 19, 81– 94. Simpson, R. H., N. Frank, D. Shideler, and H. M. Johnson, 1968: Atlantic tropical disturbances of 1967. Mon. Wea. Rev., 96, 251–259. ——, ——, ——, and ——, 1969: Atlantic tropical disturbances of 1968. Mon. Wea. Rev., 97, 240–255. Tarhule, A., and P. J. Lamb, 2003: Climate research and seasonal forecasting for West Africans: Perceptions, dissemination, and use? Bull. Amer. Meteor. Soc., 84, 1741–1759. Wade, N., 1974: Sahelian drought: No victory for Western aid. Science, 185, 234–237. Walsh, J., 1984: Sahel will suffer even if rains come. Science, 224, 467–471. Ward, M. N., 1998: Diagnosis and short-lead time prediction of summer rainfall in tropical North Africa on interannual and multidecadal timescales. J. Climate, 11, 3167–3191. Wilks, D. S., 1995: Statistical Methods in the Atmospheric Sciences. Academic Press, 467 pp.