a. Topography and land cover

The Southern Super Site (SSS) covers an area approximately 10 km × 10 km around 13.2°N, 2.3°E, at the center of the southwest quadrant of the HAPEX 1° square. The topography and vegetation cover of the region are shown in Fig. 1. The SSS lies about 5 km to the west of the Niger River, on a very gentle slope. To the west of the SSS, at a height of about 60 m above the river, laterite plateaux dominate. These are covered by “tiger bush,” comprising bands 10–30 m wide of dense woody perennials, separated by strips of completely bare, crusted soil. Elsewhere, the soil is predominantly sandy and largely cultivated with millet. Areas of natural fallow savannah are interspersed, with scattered shrubs found among herbs, grasses, and occasional trees. A full description of the SSS, the different vegetation types, and details of the various measurement systems is given by Wallace et al. (1994). The data used in this study are described briefly below.

b. Available data

The SSS comprises three subsites about 5 km apart representing the predominant land cover types. At each of these sites there are automatic weather stations (AWS). These provide measurements of temperature, humidity, wind speed, and direction, incoming and outgoing components of shortwave radiation, and net radiation. Accompanying each of these is a Hydra eddy correlation system operated by the Institute of Hydrology (IH). These measure turbulent fluxes of heat, moisture, and momentum. The EPSAT-Niger network incorporates five rain gauges at the SSS, including one at each subsite, one to the north (Diokoti) and one to the south (Djakindji). Elsewhere in the HAPEX square, rain gauges are found at roughly 12.5-km intervals.

On certain days during the IOP, radiosondes were launched every 2 h at the savannah or tiger bush sites (Dolman et al. 1997). Planetary boundary layer measurements are also available from a number of flights by the Merlin aircraft (Saïd et al. 1997). On several days, the flight path crossed the edge of the SSS at different heights. Observations from these flights include temperature, humidity, horizontal wind components, and turbulent fluxes.

c. Rainfall conditions over the SSS

Of the 107 sites in the EPSAT study area, the highest total rainfall of 782 mm for the 1992 wet season was recorded at Diokoti (Lebel et al. 1997). This compares with an average across the HAPEX square of 513 mm. The remaining four rain gauges around the SSS revealed sharply decreasing rainfall accumulations moving south, with only 507 mm at Djakindji. Rain gauges operated by IH at the three SSS subsites provide independent confirmation of the strong rainfall gradient across the area (S. Gaze 1995, personal communication).

The spatial extent of this rainfall variability can be seen in Fig. 2. This shows isohyets across the southwest quadrant of the HAPEX square for the periods 1 May–30 July (days 122–212) and 31 July–18 September (days 213–262). Contours have been calculated from the total EPSAT-Niger rain gauge measurements for the period using bilinear interpolation. It should be noted that considerable rainfall variability occurs on scales of less than 10 km, so that the accuracy of the isohyets is questionable away from the denser network at the SSS. The isohyets show that in the 3-month period prior to 31 July, gradients of accumulated rainfall were small around the SSS. However, in the final seven weeks of the wet season, the EPSAT-Niger network resolves an extreme mesoscale rainfall maximum across the SSS.

The temporal evolution of the extreme rainfall gradient is presented in Fig. 3. This compares the average of the daily rainfall totals at the two northern sites (Diokoti and savannah) and the two southern gauges (tiger bush and Djakindji). From this time series, it is clear that the rainfall anomaly in Fig. 2b results from a persistent latitudinal gradient across the SSS in almost every event. The difference does not arise from a higher number of events in the north, but rather more intense events. Of the 16 days in this period when rain of 10 mm or more is recorded on one of the gauges, only 2 days show a reversal of the north–south gradient.

Accumulated rainfall totals at the five SSS gauges are given in Table 1 from 1991, and for the two periods illustrated in Fig. 2. The higher rain in the north of the SSS in August and September 1992 contrasts with small gradients in the opposite sense in 1991 and earlier in 1992. The available data from these five gauges show that excluding the final 7 weeks of the 1992 wet season, the total rain for the years 1990–94 in the north is within 1.5% of the total rain in the south. This compares with a difference of over 100% between the northern and southern gauges in August and September 1992. It appears that local features, such as the proximity of the river and the distribution of plateaux do not by themselves favor enhanced rainfall around Diokoti and the savannah site.

The persistent rainfall anomalies during this period hint at the presence of a surface feedback operating at the end of the 1992 wet season. This paper assesses the potential for such a feedback using measurements from HAPEX-Sahel. The following sections investigate the influence of antecedent rainfall variability across the SSS on the surface and PBL, while possible feedback mechanisms in subsequent rainfall events are proposed in section 6.

a. Soil moisture

The influence of the rainfall gradient on soil moisture during the IOP is illustrated in Fig. 4. This relates differences in rainfall to soil moisture stored in the top 150 cm at the savannah and millet sites. When measurements start on 19 August (day 232), the soil at the savannah is more than 50 mm wetter than at the millet site. In subsequent years, following more uniform rainfall at the beginning of August, the soil moisture is quite similar at the two sites (Gaze 1997). The difference in soil moisture at the start of the IOP is therefore due to rainfall variability rather than the details of the soil or vegetation at the two sites.

Differences in soil moisture at the two sites of more than 30 mm persist throughout the remainder of the wet season due to cumulative differences in rainfall of 110 mm. Compared to the extreme contrast of 284 mm between Diokoti and Djakindji, the difference in rainfall between these two sites is small. Given the same surface characteristics across the SSS, the difference in soil moisture between Diokoti and Djakindji would be even larger than that apparent in Fig. 4b.

b. Surface fluxes

The surface exchanges of water and heat differ at the three flux sites over the SSS in part because of the different physical and physiological properties of the land cover types and soils (Gash et al. 1997). However, it is important to remember that all three surface types have considerable areas of bare soil. Of interest here is the sensitivity of surface fluxes to rainfall and soil moisture during the IOP.

The sum of the daytime average sensible and latent heat fluxes, the available energy, is presented in Fig. 5a for each of the three cover types. The data are normalized by the incoming solar radiation to reduce the influence of cloud cover on the surface energy budget. This illustrates well how the available energy responds to rainfall on timescales of several days. While the vegetation types exhibit characteristic differences throughout the IOP, the available energy increases dramatically for all three surfaces after rainfall, for example, after days 234, 255, and 256.

The sensitivity of available energy to rainfall on timescales of several days is due to the large percentage of bare soil in the Sahel. Near-surface moisture in the days following rainfall reduces the albedo (Allen et al. 1994). In addition, high rates of bare soil evaporation (Wallace and Holwill 1997) keep the surface relatively cool, leading to lower longwave emission and higher available energy. The magnitudes and timescales of these changes are dependent on complex features of the soil and vegetation at each site.

The daytime averaged evaporative fraction, the evaporation normalized by the available energy, is presented in Fig. 5b for the three sites. Very high evaporation is found in the day or two following rainfall, primarily coming from bare soil. Over dry periods of a few days, the evaporative fraction falls at all three flux sites due to decreasing near-surface soil moisture. Examples of this are apparent at the tiger bush site prior to day 235, at all three sites after the rain on day 244, and at the start of the dry down around day 260. Determining the influence of individual factors on surface fluxes in Fig. 5 is difficult because of contrasts in soil, vegetation, and rainfall between the sites. However, the consistently high wet season evaporative fraction for the savannah is likely to be due to both the relatively low bare soil fraction of the vegetation type, and the higher antecedent rainfall at this site.

c. Leaf area

On longer timescales, the development of Sahelian vegetation is intimately linked to soil moisture. Response of leaf area to the mesoscale rainfall gradient over the SSS can be examined with maps of modified soil adjusted vegetation index (MSAVI; Qi et al. 1994). These have been derived from Advanced Very High Resolution Radiometer images through the IOP (Kerr et al. 1993).

Figure 6a shows contours of maximum MSAVI for the period from day 220 to 229 (7–16 August). Marked gradients in leaf area have developed that were not apparent earlier in the growing season. The MSAVI images are sensitive to both rainfall and land cover type and reveal a maximum around the irrigated river valley. A band of higher MSAVI extends westward from the valley across the SSS. Figure 6b shows the difference in MSAVI between 1992 and 1991 and is more sensitive to interannual variability in rainfall. This indicates a longitudinal band of relatively well-developed vegetation between 13.2 ° and 13.3°N corresponding roughly to the 1992 rainfall pattern. Elsewhere, the leaf area anomalies and isohyets show less correlation. This is in part due to the contrast in resolution of the datasets.

d. Summary

Both available energy and evaporation are sensitive to antecedent rainfall at a range of timescales for the three predominant cover types around the SSS. It is proposed that at the mesoscale, rainfall variability around the SSS is the dominant factor in determining gradients of surface fluxes. Areas of high antecedent rainfall have higher evaporation and available energy than areas of low rainfall. Surface variability across the SSS is likely to be larger in periods when no rain has fallen for several days in the south of the area. If rainfall-induced surface flux gradients are large enough, this should generate variability in PBL properties across the SSS. This proposition is tested in the following sections with an analysis of atmospheric measurements.

a. Variability throughout the IOP

Measurements of temperature and humidity are available throughout the IOP from the AWS at the tiger bush and savannah sites. Analysis of these data in terms of surface-induced mesoscale variability can provide an understanding of the effects of antecedent rainfall patterns on the PBL at this scale.

The AWS are at heights of 14 and 7 m, respectively, and have been interpolated in the vertical to 9.5 m using eddy correlation data and Monin–Obukhov similarity theory. Figure 7 presents the potential temperatures and humidities in terms of daily averaged differences between the savannah and tiger bush sites through the IOP. This shows that the savannah is typically about 0.3 g kg⁻¹ moister than the tiger bush site but at a similar temperature. The differences are largest during the dry periods prior to days 235 and 256, and decrease rapidly following widespread rain.

Surface-layer differences between the savannah and tiger bush are analyzed in Fig. 8 during the dry periods up to days 234 and 255. This shows humidity and potential temperature contrasts at the savannah site as a function of time of day and wind direction. A pronounced diurnal cycle is evident in Fig. 8a, with the largest differences between the sites during late afternoon and early evening. During wetter periods (not shown), the differences are considerably smaller and there is no diurnal cycle. According to Fig. 8b, the largest hourly contrasts are found when the surface-layer wind blows from the south west.

Independent measurements of temperature and humidity are available from the Hydra eddy correlation systems at the sites. These provide confirmation of both the magnitude and the evolution of the savannah site anomalies. However, because of the nature of the comparison, the accuracy of the measured anomalies remains at ±0.2 g kg⁻¹ and ±0.2°C.

b. Interpretation

The measurements appear to support the proposition that antecedent rainfall generates mesoscale variability across the SSS in the PBL. Periods during the IOP when surface-layer anomalies are large coincide with contrasts in evaporative fraction between the sites (Fig. 5b), for example, prior to rain on days 235 and 256. As discussed in section 3b, these are related to differences in soil moisture stress on evaporation.

From certain wind directions, differences in upwind soil moisture between the savannah and tiger bush sites are implied by the isohyets in Fig. 2b. Mesoscale advection of upwind PBL humidity appears to influence the measured anomalies in Fig. 8b. When the surface layer wind is from the south and west, the area upwind of the tiger bush site is markedly drier than upwind of the savannah site in Fig. 2b. When the wind is more northerly, both sites are downwind of the wet patch, and measured differences are generally smaller.

Surface-induced PBL variability should exhibit a pronounced diurnal cycle. At night, when surface fluxes are small, horizontal advection will degrade mesoscale gradients in the PBL. Following the breakup of the nocturnal inversion in the morning, however, horizontal gradients over the depth of the PBL are likely to increase given fixed wind conditions. In the afternoon, the surface layer is closely coupled to the well-mixed PBL. Afternoon contrasts at 9.5 m thus indicate marked horizontal gradients across the SSS extending throughout the entire PBL. The use of surface-layer differences as a proxy for PBL gradients is less accurate in early morning and around midday, due to the development of a nocturnal stable layer and large surface fluxes, respectively.

All of these features suggest that the measured anomalies are not due to inherent errors in the comparison, but support the proposition that antecedent rainfall influences variability in the PBL at this scale. They suggest that a mesoscale moisture gradient across the SSS persisted throughout the IOP. This runs counter to the smaller, synoptic-scale decrease of humidity with latitude characteristic of the Sahel. In the next section, aircraft measurements in the well-mixed PBL are analyzed to assess the vertical and horizontal extent of these anomalies.

a. Flights over the SSS

On a number of days during the IOP, the Merlin aircraft flew over part of the SSS. Data from these flights have been analyzed to ascertain the magnitude and spatial scale of gradients in PBL properties. Data from three flights are presented here, at midday on 20 August (day 233), and between 1015 and 1115 UTC 9 September (day 253), and 2 October (day 276). The flights were undertaken at constant pressure altitude and at various levels. Leg 1 corresponds to a height of 50 m above the mean relief, leg 2 (600 m) is close to the boundary layer top, while leg 3 (300 m) is in the middle of the late-morning well-mixed layer. The PBL wind on 20 August is a strong westerly (∼7 m s⁻¹), while modest westerly flow (∼4 m s⁻¹) is found on 2 October. On 9 September, however, the wind is light (∼2 m s⁻¹) and more northwesterly. Conditions were relatively cloud-free and represent periods when the surface around the SSS was quite variable. Isohyets of accumulated rainfall over the previous three weeks are shown in Fig. 9, along with flight paths. The aircraft measurements of potential temperature and humidity are presented in terms of local deviations from the average along the flight leg (Fig. 10).

On 20 August (Figs. 10a and 10d), notable gradients of temperature and humidity are found when flying from b to a. While the potential temperature drops by 1°C in 60 km during the flight over the northern edge of the SSS, the PBL humidity increases by 1.4 g kg⁻¹ over a similar distance.

The flights on 2 October (Figs. 10c and 10f) are qualitatively similar to that on 20 August. Over the southeastern half of the legs, humidity increases while the PBL temperature drops. The gradients between the eastern and western zones appear to be a little smaller than on 20 August.

The largest variability in the aircraft measurements between a and b occur during the three legs under calm conditions on 9 September (Figs. 10b and 10e). On the first leg, at 50 m above the surface, alternate warm and cool regions are encountered, accompanied by opposing humidity anomalies. Over a distance of 30 km, between 13.45°N, 2.0°E and 13.33°N, 2.2°E, the measured humidity increases by approximately 2 g kg⁻¹. On the return flight at 600 m, high temperatures and low humidities are recorded over this same area. In the final flight, at an intermediate height, the trends tend to be of the same sign as those in leg 1 but smaller in magnitude.

These flights can be interpreted with the aid of vertical profiles from the radiosondes launched at 0855 and 1134 UTC that morning at the savannah site (Fig. 11). In the earlier ascent, the well-mixed PBL is capped by an inversion at 350 m. By 1134, the PBL has grown to a height of 650 m. Considerable vertical gradients of humidity and potential temperature exist above the capping inversion. Flying at a height of 600 m around 1055, the aircraft on leg 2 is close to the inversion. The observed drop in specific humidity of −2 g kg⁻¹ and increase in potential temperature of 1°–2°C to the northwest of the SSS are typical of changes in the vertical profiles over depths of approximately 50 m near the inversion. This suggests that between 13.38°N, 2.1°E and 13.33°N, 2.2°E, the plane is actually flying through a locally lower inversion. This interpretation is supported by the measurements on legs 1 and 3, which indicate the area to be notably moister and cooler at low levels than elsewhere.

Analysis of other legs flown on these days does not reveal comparable humidity gradients oriented in the direction of flight. This suggests that the features are not large scale, but are due to the presence of mesoscale variability in the sources of low-level heat and moisture. Aircraft estimates of surface latent and sensible heat fluxes, available for the low-level legs on 9 September and 2 October, can shed some light on the source of the variability. The turbulence data have been reanalyzed to assess whether changes in surface flux regime are resolved along the length of the low-level legs. The results, shown in Table 2, indicate that on both days, gradients in PBL variables correspond closely to independently calculated transitions in measured fluxes. For example, on 2 October, the area southeast of 13.35°N, 2.18°E is marked by an increase in evaporative fraction and coincides with the transition to moister and cooler PBL conditions. This suggests that the gradients shown in Fig. 10 are due to variability in surface fluxes.

Only over the eastern half of the 9 September flight does this correspondence between measured flux and state variable break down. Here, the pronounced moist and cool region between 2.1° and 2.2°E in Figs. 10b and 10e coincides with measurements of low evaporation and low available energy. The dry PBL at the eastern end of the leg, however, appears to be marked by very high evaporation and available energy. Possible reasons for this discrepancy will be discussed in the next section.

The aircraft data are consistent with the proposition that mesoscale flux gradients are strongly influenced by antecedent rainfall around the SSS. The gradients in the PBL on 20 August and 2 October generally follow the isohyets in Fig. 9. On 20 August, the PBL is about 0.5 g kg⁻¹ moister and 0.5°C cooler than the leg average over the high rainfall band around 13.3°N. By 2 October, no rainfall has fallen for over 2 weeks. A similar PBL moisture anomaly is apparent as the aircraft crosses the high seasonal rainfall streak evident in Fig. 2b.

The agreement between the isohyets and PBL gradients is poorer on 9 September than for the two windier flights. Lower wind speed implies increased residence time for air passing over distinct landscape features. This shortens the length scale at which PBL features are organized. In these conditions, surface variability on scales of about 10 km appears to induce considerable variability in the PBL. The only area along the flight path that the EPSAT-Niger rain gauge network resolves at a scale of less than 10 km is the SSS. Here, the aircraft measures a drop in the PBL humidity as it approaches the drier millet site.

b. Planetary boundary layer variability on 9 September

The low wind conditions on 9 September make it an ideal day to examine PBL heterogeneity around the SSS in greater detail. Data from two further flight legs across the south of the SSS are shown in Fig. 12. These reveal the presence of a dry and warm region to the northeast of 13.05°N, 2.28°E on leg 4 at 60 m and to the east of 13.17°N, 2.2°E on leg 5 (500 m). This area includes the particularly dry tiger bush and Djakindji rain gauges.

Horizontal temperature gradients of 0.7°C over 20 km are found approaching the millet site on leg 1 and of 0.9°C over 15 km to the south of the SSS on leg 4. These temperature changes are plotted alongside the wind components parallel and perpendicular to the direction of the flight paths on the low-level legs 1 and 4 in Fig. 13. On both of these legs, there is a strong positive correlation between the along-path velocity and the temperature gradient in the direction of flight. Changes in the cross-path velocities also correspond to regions of large thermal gradients. The increases in low-level wind approaching the warm sector hint at the presence of a thermally driven boundary layer circulation, although direct measurements of vertical velocity are not available to confirm this feature.

Independently calculated transitions of surface flux regime along both low-level flight legs (1 and 4) coincide with PBL gradients. This implies that surface flux variability is responsible for atmospheric gradients. Airborne estimates show increases in sensible heat flux of about 10–25 W m⁻² over the warm sectors. In addition, both of the dry sector surface flux sites (millet and tiger bush) have relatively low evaporative fraction. This contrasts with the savannah site, which has received markedly more rainfall and maintained higher soil moisture than the south and east of the SSS. However, because of differences in land cover, the influence of rainfall on sensible heat flux variability cannot be isolated.

Analysis of the patch size at which heterogeneous terrain can induce a mesoscale circulation has been presented by several authors (e.g., Segal and Arritt 1992; Doran et al. 1995). These simple studies suggest that, with a wind speed of 2 m s⁻¹, contrasts in sensible heat flux of 20 W m⁻² organized at length scales of 15 km can induce boundary layer circulations. From the three available flight lines on 9 September, the warm sector appears to stretch from about 13.25° to 13.05°N, and 2.2° to 2.35°E (∼20 km × 15 km). The estimates of surface fluxes and length scales around the SSS thus satisfy the criteria for mesoscale heterogeneity to induce circulations.

The observations are qualitatively similar to other aircraft studies of nonclassical mesoscale circulations above irrigated areas (Segal et al. 1989; Mahrt et al. 1994) and snow cover (Segal et al. 1991). In these cases, the differential heating of the PBL is more well-defined spatially but, in the case of the study by Segal et al. (1989), was complicated by the presence of orographic flows. Modest orographic features are present over this flight path, in particular the Niger River valley. However, the low-level increases in velocity in legs 1 and 4 are down slight slopes (∼0.01) of the river valley. A classical valley flow would lead to upslope acceleration in these cases.

The apparent discrepancy between aircraft estimates of high evaporation southeast of 13.28°N, 2.27°E on 9 September (Table 2) and low humidity in this sector can be interpreted in terms of the mesoscale circulation. As discussed by Mahrt et al. (1994), in regions of mesoscale convergence, aircraft flux measurements even a few tens of meters from the ground may not be representative of surface values. From Fig. 13a, it is clear that the independently calculated flux transition at 2.18°E also coincides with implied changes in convergence. In this case, surface-induced mesoscale eddies may lead to the dramatic increase in measured latent heat flux (Table 2) in an apparently dry and warm sector.

a. Rainfall and moist static energy anomalies

Previous sections have established the presence of rainfall-induced anomalies in the PBL across the SSS. In this section, we investigate the links between atmospheric variability and subsequent rainfall patterns.

A useful property for examining the energy available to a parcel in a convective storm is the moist static energy h (=c_pT + λq + gz). Moist static energy anomalies across the SSS are presented in Fig. 14 during the IOP. These are found from the AWS temperature and humidity differences in Fig. 7. A moist static energy contrast of about 0.7 kJ kg⁻¹ persists throughout the wet season between the savannah and tiger bush sites. As in Fig. 7, there is a close relationship between daily averaged differences and antecedent rainfall. Drier periods, when evaporation at the tiger bush site appears to be water stressed, correlate well with larger contrasts in moist static energy.

The diurnal variability in the rainfall gradient is illustrated in Fig. 15. Here, all 15 events of 10 mm or more during August and September have been classified according to the time of first rainfall at the SSS. Averaged over the SSS, the most rain falls during the five events in the evening period (2000 to 0100 UTC). Only 25 mm is produced by the two midmorning events.

The differences between the northern and southern gauges also show a diurnal variation. As a percentage of the average rainfall from the four gauges, the rainfall in the north exceeds that in the south by over 100% in the afternoon period. This decreases to 85% for the evening events and drops considerably after 0200 UTC. The phase of this diurnal cycle corresponds closely to that found in Fig. 7b. In the afternoon, prestorm anomalies of humidity and moist static energy are large and likely to extend through a deep, well-mixed PBL. Compared to its average across the SSS, rainfall variability is also largest at this time of day, suggesting a causal link.

In the next sections, we examine possible PBL influences on rainfall. Unfortunately, observations of the prestorm PBL and its variability at this scale are very limited. Two contrasting case studies are presented using aircraft data from the previous days and surface observations.

b. Squall line on 21 August

Of the 15 major events during August and September (Fig. 3), three large-scale storms produced differences in rainfall between the north and south of the SSS in excess of 30 mm, on days 221, 228, and 234. All three of these events occurred during relatively dry periods at the tiger bush site. The prestorm PBL may have exhibited marked gradients in humidity as a result. This section investigates the third event on 21 August (day 234, the start of the IOP), when some estimates of prestorm PBL contrasts are available.

The aircraft data from 20 August reveal a positive PBL moisture anomaly around the north of the SSS (Fig. 10a). No rain fell between then and the evening of 21 August, so surface flux patterns were probably similar. Evidence that the PBL moisture gradient observed by aircraft on 20 August remained the following day comes from the AWS measurements (Fig. 7). Daily average specific humidity differences of 0.5 g kg⁻¹ between the savannah and tiger bush sites were recorded on both days. By late afternoon on 21 August, the savannah site was 1 g kg⁻¹ moister than the tiger bush site, while remaining less than 0.2°C cooler. This gives almost 2 kJ kg⁻¹ excess low-level moist static energy at the savannah site.

Isohyets for the storm on 21 August are shown in Fig. 16a. A westward-moving squall line passed the SSS around 2300 UTC, producing 65 mm of rain at Diokoti, compared to 27 mm at Djakindji. The disturbance appears as a band of deep cloud extending 600 km from south to north on Meteosat images. Although such events bring heavy rain over a large area, there is appreciable smaller-scale structure within any squall line (e.g., Houze 1977; Chong et al. 1987). The disturbance is generally made up of short-lived convective cells found on length scales typically ranging from 5 to 25 km. New cells are generated at the leading edge of the squall line with conditionally unstable, near-surface air lifted to heights of 10–20 km.

The passage of the squall line can be seen from a time series of AWS data at the SSS (Fig. 16b). The gust front of the storm reaches the SSS around 2230 UTC, and is marked by a drop in temperature of 6°C and a peak in the surface-layer wind speed to over 10 m s⁻¹. In this case, there is no rain at the SSS for a further 30–40 min, implying a large tilt in the leading edge of the squall line (e.g., LeMone et al. 1984). About 1 h after the passage of the gust front, rainfall rates at the savannah site exceed 60 mm h⁻¹, compared to about 20 mm h⁻¹ at the tiger bush site. This suggests that the northern end of the SSS is beneath a deep convective cell, while convection is less active farther south. After 2400 UTC, the rainfall rates decrease and differ little across the SSS as the anvil of the storm passes overhead. The gradient in rainfall in this storm, as is typical of events throughout the period, is due to a local intensification of convection behind the leading edge of the squall line.

While the evolution of a squall line is determined by processes at a large scale, individual cumulonimbi may be sensitive to near-surface conditions on convective length scales. The horizontal and vertical extent of the prestorm moisture anomaly around the SSS is not clear, nor is the sensitivity of deep convection to anomalies of this magnitude. If one assumed that horizontal anomalies in the low-level inflow were as large as those measured at 9.5 m (δq ≈ 1 g kg⁻¹ and δθ ≈ −0.2°), differences in the wet-bulb potential temperature above cloud base between the two sites would be approximately 1°C. From a simple tephigram construction, this leads to a difference in convective available potential energy (CAPE) of about 600 J kg⁻¹, given a cloud depth of 10 km. Such a contrast would represent a significant local perturbation to a tropical squall line of typical total CAPE of 1500 J kg⁻¹. It seems plausible then that measured surface-induced prestorm anomalies are large enough to influence rainfall at convective length scales in passing storms.

Total rainfall following the passage of an intense convective cell within a squall line may exhibit several characteristics. Rainfall will be highest over the track of the cell during its active lifetime. Marked gradients perpendicular to the direction of travel are likely to be found on the length scale of the cell. Taking typical values of cell lifetime of 20–30 min, and a storm speed of 15 m s⁻¹, this streak of high rainfall may stretch some 18–27 km in the direction of travel of the squall line. In Niger, storm paths are strongly anisotropic, predominantly moving toward the west. In these conditions, if a preferred area of intensification persists over a series of storms, this should be apparent in the isohyets for that period. One might expect an east–west band of high rainfall with dimensions governed by characteristic length scales of a convective cell. This simple conceptual model is consistent with the mesoscale rainfall maximum 40 km × 20 km in Fig. 2, as partially resolved by the EPSAT network.

c. Isolated convective storm on 10 September

Most of the rain in the period studied results from the passage of large-scale disturbances such as squall lines. In only one case does an isolated storm produce a strong rainfall gradient across the SSS. This day was 10 September (day 254), where 20 mm of rain fell around 1330 UTC at the savannah site. By contrast, only the nearby Diokoti gauge (4 mm) recorded rainfall on this day at the SSS. To the north and west, 16 other EPSAT gauges received generally very light scattered rain.

An estimate of PBL variability on 10 September can be obtained from the aircraft data (section 5) on the previous day. Surface flux patterns are probably similar, as no rain fell in the interim period. The stronger wind on 10 September (3.5 m s⁻¹ at 10 m) will have smoothed out variability at the small scales found in Fig. 10b.

In the absence of low-level convergence provided by large-scale storm systems, the influence of surface flux variability on moist convection is sensitive to the depth and intensity of the capping inversion (e.g., Ek and Mahrt 1994). This is apparent from the profiles for the 1100 UTC sounding at the savannah site shown in Fig. 17, plotted on T–ϕ axes. An inversion around 900 mb caps the well-mixed PBL with stable lapse rates above. However, given a PBL mixing ratio of 17 g kg⁻¹, the atmosphere is conditionally unstable to moist adiabatic ascent to beyond the height of measurement at 6 km.

The temperature from the 1300 UTC sonde is also plotted up to about 875 mb. At this level, the sonde moved rapidly south in the gust front of an unsighted cumulonimbus. Assuming an increase in height, and weakening in strength of the inversion between 1100 UTC and 1300 UTC, estimated cloud base at the savannah site (marked “X” in Fig. 17) is within the PBL. Available surface data across the SSS at 1300 UTC reveal that the surface layer above the savannah may have been about 0.5 g kg⁻¹ moister than at the millet site, while remaining at a similar temperature. Extrapolating this gradient into the well-mixed PBL, it is clear that at the wetter savannah site, cloud base is reached earlier than over drier areas. Subsequent moist-adiabatic ascent is more likely to break through the inversion and ascend through the deep unstable layer above, resulting in localized and intense rainfall. It appears then that modest humidity anomalies of the magnitude measured throughout the IOP can have a marked influence on the stability of the atmosphere to deep convection.

d. Discussion

Surface-induced PBL anomalies around the SSS share similar temporal characteristics to the local rainfall gradients on both diurnal and daily timescales. In addition, the magnitudes of measured anomalies appear to be large enough to influence rainfall in large scale as well as isolated storms. To explain the persistent and extreme precipitation patterns over the SSS, we propose that gradients in antecedent rainfall produce a positive feedback on rainfall via surface processes. Enhanced evaporation and available energy in wetter areas induce local maxima in PBL moist static energy, which in turn enhance rainfall.

The efficiency of this proposed feedback relies on both surface and atmospheric features typical of the Sahel. The high bare soil fraction of the vegetation leads to high spatial variability in evaporation from antecedent rainfall patterns and smaller gradients in sensible heat flux. In terms of moist static energy, this implies that greater humidity in wetter areas will outweigh cooler temperatures and produce pronounced positive anomalies. This sensitivity of available energy to soil moisture is not incorporated in the simple feedback model of Ek and Mahrt (1994), for example.

To maintain a potential feedback, an important feature is the presence of low-level convergence to initiate deep convection (as found, e.g., in Barnston and Schickedanz 1984). In the feedback hypothesis of Anthes (1984), this was provided by organized mesoscale circulations induced by bands of irrigated land some 50–100 km across. By contrast, in the majority of events in this study, convergence is provided by the frequent passage of squall lines, a key characteristic of the Sahelian wet season.

The spatial scale of the proposed feedback contrasts with other studies over larger areas (e.g., Anthes 1984; Oglesby and Erickson 1989). The size of the resulting wet area appears to be influenced by dominant length scales in large-scale disturbances. The orientation of the high rainfall band is related to the anisotropy of storm travel. Low-level wind shear may be influential on the rainfall feedback via advection of PBL properties toward incoming disturbances and through tilting squall-line updrafts.

Testing of this feedback hypothesis requires a series of numerical modeling experiments using realistic surface boundary conditions. These should investigate the role of antecedent rainfall and land cover on the development and scale of prestorm atmospheric anomalies. Questions concerning the importance of PBL variability, the proximity of the river, and orographic features on convective events can be answered only by numerical modeling. Mesoscale simulations at a larger scale (Taylor 1996) have shown a sensitivity in parameterized convection and precipitation to realistic prestorm anomalies on a single day in the wet season. However, only a model that resolves cloud-scale processes within a squall line environment can provide a thorough understanding of potential feedbacks and reveal their sensitivity to atmospheric features.