a water balance study of the upper white nile basin flows...

Hydrological Sciences-Journal-des Sciences Hydrologiques, 46(2) April 2001 301

A water balance study of the upper White Nile basin flows in the late nineteenth century

EMMA L. TATE, KEVIN J. SENE Centre for Ecology and Hydrology, Wallingford, Oxfordshire OX10 8BB, UK

JOHN V. SUTCLIFFE Heath Barton, Manor Road, Goring-on-Thames, Oxfordshire RG8 9EH, UK

Abstract A water balance model of the Nile is described, working in terms of dry season flows. The main use of the modelling studies was to extend flows back to the start of observations at Aswan in 1869, and hence to estimate the levels of Lake Victoria during the high flow event of the 1870s. New estimates of Lake Victoria levels and outflows for the period 1870-1895 are deduced, extending the length of the observed data for Lake Victoria by some 25%. Crucially, the analysis includes the major flood event of 1878, which historical evidence suggests was the highest since 1860 or earlier. The modelling results are compared with previous estimates and a detailed examination of historical evidence. Some prospects for future levels are considered, taking into account the reconstructed levels of the late nineteenth century.

Key words Lake Victoria outflows; reconstructed levels; water balance; river flow

Une étude du bilan hydrologique du bassin du haut Nil Blanc pendant la dernière partie du dix-neuvième siècle Résumé Un modèle du bilan hydrologique du bassin du Nil est décrit, en fonction des écoulements de la saison sèche. Les études de modélisation ont été principalement orientées vers l'extension des données de débit jusqu'au début des observations à Assouan en 1869, et, en amont, vers une estimation du niveau du lac Victoria dans la période de crue des années 1870. De nouvelles évaluations des niveaux d'eau et des flux sortant du Lac Victoria ont été déduites pour la période 1870-1895, rallongeant ainsi de 25% environ la série des données observées. Les analyses incluent en particulier l'événement de crue majeur de 1878, événement considéré sur le plan historique comme le plus important depuis 1860 (voire auparavant). Les résultats de la modélisation sont comparés avec des évaluations antérieures et avec une étude historique détaillée. Des prédictions des niveaux d'eau futurs ont été réalisées, prenant en compte les niveaux reconstitués de la fin du dix-neuvième siècle.

Mots clefs flux sortant du Lac Victoria; niveaux reconstitués; bilan hydrique; débit

INTRODUCTION

The White Nile is one of the; two major rivers which join at Khartoum to form the River Nile, the other being the Blue Nile (Fig. 1). Numerous hydrological studies (e.g. Hurst, 1952; Sutcliffe & Parks, 1999) have shown that the base flows in the White Nile are dominated by the influence of the many lakes and swamps in the upper basin, and in particular by Lake Victoria which is the main source of White Nile flows.

Considering Lake Victoria's dominant effect not only on White Nile flows, but also on hydropower production in the region, this paper attempts to augment the understanding of the variability of Lake Victoria outflows. This is achieved through extension of the lake outflow time series by some 25%, back to 1870, incorporating the

Open for discussion until 1 October 2001

302 Emma L. Tate et al.

40°E

Key to dam location numbers

1 Aswan

2 JebelAulia

3 Khashm el Girba

4 Sennar

5 Roseires

6 Owen Falls

N C O N G O

Fig. 1 Map of the Nile basin.

major flood event of 1878. The time series extension is achieved using a water balance model, which employs linear regressions to create the full time series back to 1870.

Lake level and rainfall observations within the Lake Victoria basin first started in 1896, and until the 1960s lake levels and outflows were thought to vary within a

A water balance study of the upper White Nile basin flows in the late nineteenth century 303

relatively small range. However, in the period 1961-1964, following widespread heavy rains in East Africa, the level of Lake Victoria rose by over 2.5 m and more than doubled the lake outflow and the area of the Sudd swamps downstream. This event occurred shortly after the completion at the lake outlet of the Owen Falls Dam, which supplies electricity to Uganda and Kenya, and raised questions about the future hydropower potential of the Nile below the lake, and whether the spillway capacity needed to be raised. Subsequent studies (e.g. Piper et al, 1986; Sene & Plinston, 1994; Sene, 2000; Yin & Nicholson, 1998) have demonstrated the extreme sensitivity of the lake balance to variations of lake rainfall and climate, and have highlighted the need for caution in predicting future lake levels.

Since that time, lake levels have followed three decades of gradual decline, punctuated by short-lived increases in the late 1970s and in 1998 when values approached the earlier maximum. The most recent event followed heavy rainfall during the 1998 El Nino event. This raises the interesting question of whether such events will become more frequent, and the problem of estimating the return periods of such events from the historical record. However, on the basis of so few events, it is difficult to know whether sudden rises of this magnitude form part of the natural regime of the lake, or whether they are a more recent phenomenon. This is partly because the long response time of Lake Victoria, which is of the order of a decade, complicates the response of lake outflows to variations of rainfall, evaporation and inflows in the lake basin (Sene & Plinston, 1994). It is useful, therefore, to search for additional evidence of lake level variations before the start of observations.

Based on anecdotal evidence, mainly collected by Lyons (Garstin, 1904, Appendix III; Lyons, 1906), previous studies have suggested that levels comparable to those of the 1960s may have been reached in 1875-1895, shortly before records began (Piper et al, 1986; Nicholson et al., 2000a). Attempts have been made to reconstruct the time series of lake levels during this period by correlation with low flow records at the key gauging station at Aswan in Egypt, where records began in 1869 (Flohn & Burkhardt, 1985) and by pooling evidence from travellers' accounts, missionaries and others in the region at the time (e.g. Nicholson, 1998). The aim of this paper is to take these reconstructions one stage further by means of a stochastic water balance model, and then to assess the implications, for estimating future lake levels, of combining this new record with the existing observed record.

THE STUDY AREA

Hydrology of the Nile basin

To reconstruct Lake Victoria levels before observations started in 1896, it is necessary to use flow data for stations downstream and this requires an understanding of the basin hydrology. The hydrology has been reviewed many times and only the key features will be described here, based largely on the accounts of Hurst (1952) and Sutcliffe & Parks (1999). Figure 1 shows the main hydrological features and gauging stations within the basin.

In the upper White Nile basin, the main contribution to river flows arises from the effects of two rainfall seasons on Lake Victoria (March-May and October-December).


The lake outlet is at Jinja in Uganda, where outflows were controlled naturally at the Ripon Falls until 1951. After the Owen Falls hydropower dam was built in 1951-1953, the release rales were designed to mimic the natural outflow. Lake Kyoga, some 100 km downstream, is responsible for a small evaporative loss in drier periods but for gains in the wetter period following 1961-1964. Lake Albert, some 130 km downstream, receives local inflows including the River Semliki, draining two smaller lakes (Edward and George) which lie close to the headwater tributaries of Lake Victoria. The local inflow is balanced by lake evaporation in drier periods, but provides a significant contribution in wetter periods. Between Lake Albert and the Sudd swamps, the rainfall occurs in a single season between April and October, and highly seasonal runoff occurs in the "torrents" which contributed about 15% of the total flow at Mongalla before 1960. This flow occurs mainly between May and November, and is responsible for the seasonal flooding in the Sudd region.

The next major influence on flows is the Sudd, where high flows exceed channel capacities and result in widespread inundation. The Sudd is a large area of swamps and seasonally-flooded grassland whose area varies with inflow and has ranged from less than 5000 km2 to over 30 000 km2 between 1905 and 1980. Evaporation over inundated areas is high, and on an annual basis outflows are roughly half the inflows and vary little seasonally or from year to year; it follows that losses are proportionally greater in periods of high inflows. The flows of the White Nile return to seasonal variability and higher totals with the contribution of the seasonal River Sobat, which on average almost doubles the White Nile flow. The River Sobat derives most of its runoff from the Ethiopian mountains, via the River Baro, and occasional contributions from the River Pibor, which drains a wide area to the south. Downstream of the Sobat confluence, the White Nile receives no significant inflow before meeting the Blue Nile at Khartoum. Both flood flows and dry season flows in the lower White Nile have been affected by the construction in 1934-1937 of the Jebel Aulia Dam some 40 km above Khartoum.

A notable feature of the White Nile is that the seasonal variability of flows is highly damped by lake and swamp storage within the basin; the coefficient of variation (CV) of monthly flows immediately below the Sudd (Malakal minus Sobat flows) is approximately 0.3. By contrast, Blue Nile flows are much more variable (CV of monthly flows at Khartoum is approximately 1.3) with more than 80% of flows occurring in the single flood season (June-October). Artificial influences on dry season flows in the Blue Nile arise from the seasonal irrigation/hydropower reservoirs at Sennar and Roseires, which became operational in 1925 and 1966 respectively. Below Khartoum, the Blue and White Niles join to form the main Nile. On an annual basis, the Blue Nile contributed about 65% of the combined flow until 1961; this then decreased to about 55% mainly as a result of the rise in Lake Victoria outflows since that time, but has recently returned to around 60% with the increase in Blue Nile flows since the late 1980s.

Some 1500 km downstream, the Aswan High Dam completes the controlled section of the Nile. Its predecessor, built in 1902, ended the natural regime of the Nile down to its outlet below Cairo. In the reach down to Aswan, there are losses due to evaporation, and abstractions for irrigation have increased significantly since the 1950s. The only significant inflow is from the River Atbara, which contributed about 14% of the total flow before the construction of the Khashm el Girba Dam in 1960-


1964. Previously, nearly all of the runoff from the River Atbara occurred between July and November, with the river usually dry from February to May.

Availability of data

Figure 1 shows the main long-term flow gauging stations in the Nile basin; note that many of the data used in this study have been taken from The Nile Basin, volumes III and IV (Hurst & Phillips, 1933a, b), held at CEH Wallingford. Lake Victoria levels were observed at various sites at Entebbe and Jinja from 1896, with a gap in 1897-1898; early records were converted to equivalent series at Jinja near the lake outlet. Level measurements of lakes Kyoga and Albert started in 1912 and 1904 respectively. Flow measurement in the White Nile basin began in 1905 at Mongalla, at the upstream end of the Sudd. Between Lake Victoria and Lake Kyoga, flow measurement did not start regularly until 1940 at Namasagali, but subsequent gaugings indicated the stability of the outflow rating for Lake Victoria, so an outflow record can be reconstructed from the start of level observations in 1896 until the start of operations at Owen Falls, with the measured releases used from that time. Between Lake Kyoga and Lake Albert gaugings started in 1907, but the most reliable record is at Kamdini between 1940 and 1979. At Mongalla, the increase in flows following the rise of Lake Victoria coincided with a gap in gauging, and flows in 1963-1966 appear suspect (Sutcliffe & Parks, 1999).

Downstream of the Sudd, flows at the outlet from the Sobat at Doleib Hill, and at Malakal immediately downstream on the main White Nile channel, are available from 1905 but measurements on the Sobat stopped after 1983. Outflows from the Sudd are deduced from the difference between these two sites. The White Nile flow has also been gauged at its confluence with the Blue Nile since 1911, though flows can be influenced by backwater effects from high flows in the Blue Nile.

The longest flow record on the Blue Nile is at Khartoum, where flows have been recorded since 1900, but ten-day levels near the White Nile confluence have been published for the flood season (June-October) for the period 1869-1883. There was a gap until the present gauge was established some 4 km above the confluence in 1899; by comparing levels at both gauges with levels at Aswan, Hurst & Phillips (1933a) suggested an adjustment of 3.5 m at high levels between the early and later gauges.

The historic river level records for Khartoum are potentially useful in reconstructing Lake Victoria levels before 1896. The other major flow record before 1896 is that for Aswan, which began in 1869. After the construction of the first Aswan Dam (1902) and subsequently the Aswan High Dam (1964), various versions of the flow have been published, including the "Water Arriving at Aswan" which takes account of the storage in the Aswan reservoir. An independent flow record has been published for Wadi Haifa from 1890, supplemented by Kajnarty just upstream from 1931 to 1962 and finally Dongola from 1963; the changes in location were due to successive rises in the Aswan reservoir. Both these flow records give clear but subjective indications of East African lake levels, especially the low flow periods, when White Nile flows predominate. The Roda Nilometer gauge record at Cairo (Sutcliffe & Parks, 1999), which dates from 622 AD, gives maximum and minimum levels; the record is subject to significant channel aggradation over the years, so that evidence on flood series cannot be transferred over too long a period.


WATER BALANCE MODEL

Anecdotal evidence for Lake Victoria (Piper et al., 1986) suggests that, before 1896, the highest known levels were reached in 1878, and flow records for Aswan confirm that a high flow event occurred on either, or both, of the Blue and White Niles in that year. The peak monthly flow at Aswan was in September 1878. Further, Lyons (1906, p. 103) notes a high flood on the Bahr el Jebel, and that Lado was flooded. The peak level at G'ondokoro (just upstream of Mongalla) was recorded on 22 August 1878 (Lyons, p. 109), but the level was still high in December, indicating an unusually high level of Lake Albert. This evidence is examined in more detail later, and led to the idea of attempting to reconstruct the levels of Lake Victoria during the second half of the nineteenth century.

In this water balance model, linear regressions are used to create the full time series of lake levels back to 1870, thus extending the time series by some 25%. This extended time series can be used to enhance understanding of the variability of lake levels, and assist with estimating how lake levels might look in the future. Such understanding is most useful for water resources managers and planners, especially since Lake Victoria is important for hydropower production.

It is desirable to estimate levels for the years before 1896, and, given the small amounts of data available before 1896, a linear regression approach has been adopted. Many previous studies have shown that, due to the slowly varying flow regime of the Nile, the regression approach works well, provided seasonal or annual data are used (e.g. Sutcliffe & Parks, 1999). This suggests that the influences of flow routing, ungauged inflows and losses, and lag times are small. The resulting regression relationships, and hydrological years used in the model, are summarized in Table 1.

To describe the model, the flow relationships will be considered for each reach starting at Aswan and working upstream. The approach taken in this model is to use linear regressions to extend flow series at gauged sites. In such a geographically large basin as this, climatic and hydrological regimes vary between gauging station sites, so that, for example, the dry season at Aswan is slightly different from that at Khartoum; the seasons employed are listed in Table 1. It has also been important to recognize the significant artificial influences (e.g. the construction of large dams) and hence to use data for relevant periods to develop regression relationships.

In estimating flows of the White Nile at Khartoum, with an annual approach it is not possible to estimate the effect of inflows from the Atbara with any certainty. However, during the dry season, the Atbara has no significant flow, so dry season flows at Aswan were used. Blue Nile flows are also small in the dry season, so Aswan flows at this time must arise mainly from the White Nile.

An inspection of the Khartoum and Aswan records suggests that the period when Blue Nile and Atbara contributions are small, or cease, is typically from April to June at Aswan, and March to May at Khartoum. To estimate dry season flows at Khartoum before 1896, it is necessary to relate dry season flows to the available data for the flood season (June-October), where available, or simply to assume a typical mean value. Although a recession model could possibly be developed for Khartoum flows, a simple regression model captures the main relationship between March-May flows and annual flows, whilst a second model shows a strong correlation between June-October flood flows and annual flows. Note that, for the dry season flows at Khartoum, data


3 H

°v

x> s— R3

-a

m

X )

cS

S

Î3

c p c§ (U

OT

X ! Q

J 3

S C o 2

"« > i o

E

c/3

C

&

c o c/:

« a j

co

-a .2 'C <D &H

o 4=.

c .2 o o J

^ E

o

;— O

ID

T 3 O

X ) w U

3 Ë

' w cfl <D

O

« E

8 5 o

739

_ H

CO IT)

1

\o

— H

524

NO 0 0

d

626

<N

13 7

00

0 0 NO

O N

332

cs

9 50

0

CO

d

OO OO

i n

-18

100

< N

rn

108

m

2 12

0

r^

d

' ^ H

J yea

E o O -H

Apr

. og

ical

to

um fo

r ou

s hy

drol

c3 r-«

x; c o

Sg &.&

"E *o O

~H

-Oct

.

1

toum

c3

C oui .

S NO

o 1 !

5:

en

g

in-K

hart

ou

S w <

f l l

J3 w C

9

. C

c

1 "s *o o "~H

>, C3

i.

•= is o <5

g S

kal a

t end

<

Cl

s

fi

h

Dlo

gica

l ye

i £

'*-, c3

"s *o O

' -J

an.

Feb.

ga

lla-J

inja

c o

c h E >>

~S " g o •«

o o ~-< —<

> N

43 <= 1 7

Mar

Fe

b.

kal-

Soba

t ga

lla

5 c a o

S 2 S

I c -+~J , C ' -

Q S

g E g E g

d

^

II

1965

19

33

d d o o ON ON

!

Apr

.-

9661

1906

-

!

Api

v 19

81

1905

-

7 "T X l X !

1981

19

81

1906

-19

05-

o

c W si I

o Q es 03

X > o

6 0

C5

"3 so c o

>


were only used up to 1924, because low flows since then have been influenced by the Sennar Dam (recent capacity about 0.5 km3). For annual flows, data were used up to 1965, since Roseires Dam (which is much larger than Sennar Dam, with a recent capacity of 2.4 km3) has probably influenced flood flows since then.

Flows can then be estimated at Malakal. Since 1935, dry season flows have been strongly influenced by operations at the Jebel Aulia Dam, so values from then were excluded from the regression analysis. To follow the analysis upstream, it is better to continue in terms of annual flows from this point, since both the Sobat and torrent flows have, in some years, provided large inflows during the dry season, whereas annual flows are much less variable. As on the Blue Nile, there is a reasonably consistent relationship between dry season and annual flows on the White Nile at Malakal.

To estimate flows for the Sobat, it would be desirable to represent the higher flows which seem to occur when rainfall and inflows are also high in the upper White Nile basin. These higher flows arise from the River Pibor, which runs almost parallel to the upper White Nile channel with its headwater region near to Lake Kyoga. Several possible regression relationships were investigated with inconclusive results. A weak relationship was found between annual flows in the Sobat and the net inflow to the upper White Nile between the outlet from Lake Victoria and the Sudd Swamp (i.e. Mongalla minus Lake Victoria outflows). Unlike the record for Lake Victoria outflows, this net inflow is not affected by over-year storage in the lake, and so is a reasonable indicator of annual rainfall variations in the headwaters of the Pibor.

To complete the model, it is necessary to relate the inflows of the Sudd to the outflows, and then to estimate Lake Victoria outflows. Previous studies of the Sudd have suggested roughly a two to three month lag between inflows and outflows, so hydrological years of April-March and February-January have been used for outflows and inflows respectively to try and capture this effect. On this basis, there is a reasonably linear relationship between Sudd inflows (Mongalla) and outflows (Malakal-Sobat), whilst Lake Victoria outflows are strongly related to flows at Mongalla.

Using the seven regression relationships described above (Table 1), it is possible to use the data for Aswan and Khartoum before 1896 to estimate mean annual outflows from Lake Victoria in each year from 1869 and, via the lake outflow relationship, to estimate mean annual levels. Due to the implicit relationship assumed for Sobat flows, flows upstream from Malakal cannot be estimated directly (i.e. proceeding upstream), but must first be estimated from Lake Victoria outflows following a simple algebraic rearrangement of the regression equations. Also, the model only provides a single "best" estimate in each year, with no assessment of uncertainty. To obtain a first impression of the likely uncertainty, a simple stochastic model was developed. For each regression relationship, the residuals in the relationship were assumed to be uniformly distributed across the flow range, and normally distributed with a standard deviation equal to the standard error in the relationship. The residuals at each site were also assumed to be uncorrelated with those at other sites, or with those at the same site in previous years. In some cases, there are not enough data to confirm these assumptions; however, given the available data, the stochastic model at least provides an estimate of the uncertainty associated with the estimate of flows at each site. Some example confidence limits are shown on the estimates presented later, based on 10 000 realizations of the regression model (within-year means and standard deviations typically converged to stable values after a few hundred realizations, but 10 000 realizations provide more accurate values).


RESULTS

Estimates from the model

The model was used to estimate annual mean Lake Victoria levels, and flows at other sites in the basin, for the hydrological years 1870/71-1991/92. Several tests were also made to assess the performance of the model, the first being to see how well observed flows and levels could be estimated since 1896 using only the observed dry season (April-June) flows for Aswan and flood season (June-October) flows for Khartoum. Comparison plots for Malakal and Khartoum dry season flows showed good agreement over the period of the regressions (1905-1933 and 1900-1924, respectively), although comparisons for other periods were precluded due to the major influences of Sennar and Jebel Aulia Dams from 1925 and 1934. However, the main test was to compare observed and predicted levels for Lake Victoria from the start of lake level observations in 1896.

To perform this comparison, the problem arises that, since 1934, flows in the White Nile at Khartoum have been strongly influenced by operations at Jebel Aulia Dam, particularly in the dry season. From this time, therefore, only a partial test of the model was possible, using the observed flows at Malakal as input to the White Nile component of the model, rather than the estimated flows as in the period 1896-1934. From 1934, therefore, the comparison only tests the performance of the part of the model which estimates Sobat, Sudd and lakes Kyoga and Albert flows. Figure 2(a) compares the observed and predicted levels for the period 1896-1992 on this basis. It can be seen that the model captures the sudden rise in levels from 1961 to 1964, but overestimates the decline in levels by about 0.5 m from 1964. Before 1961, agreement is good except for the short lived rise in levels of 1918, when the observed levels are overestimated by about 1.0 m. This appears to be due to the unusual combination of a sudden increase in outflows from the Lake Kyoga region, with no corresponding increase in Lake Victoria outflows.

Since the flows at Khartoum are one of the key inputs to the model, it is also worth examining this component of the model in more detail, and Fig. 2(b) shows the flood season (June-October) and annual flows (April-March) at Khartoum for the period up to 1982. Khartoum flows are not shown from 1884 to 1895 since no flood flow data were available in this period. The gradual decline in flows during the 1900s has been noted many times before (e.g. Sutcliffe & Parks, 1999) and arises both from increased losses at reservoirs and abstractions to irrigation, and a general reduction since the 1970s which seems to be linked to the general decline in rainfall throughout most of Africa over this period (e.g. Nicholson et al, 2000b; Hulme, 1990). What is perhaps more surprising is that annual flows were apparently much higher in the period 1869-1883. The recorded levels in this period provide a first check on this result. In this period, the highest monthly mean level reached at Khartoum was about 17.21 m on the Khartoum gauge in September 1878. This compares with highest values since 1900 of 16.26 m in August 1909 and 16.67 m in September 1917, implying that flows were unusually high in 1878. In an analysis of the 1988 flood at Khartoum, Walsh et al. (1994) also note that there is plenty of anecdotal evidence to back up the claim that floods at Khartoum were more frequent, and higher, during the 1870s and 1880s than at any time since records resumed in 1900; in particular, that floods and rainfall were


ed)

r*

o O

~~>

(PS

r* <u

o

m :> n <

_, '"O CD

(U

H f » i l )

m > r> <

o o o o o o o o o o o o o o o o o o o o o o o o

<0 CM T-

(jeaX/jUijOi.) M o y

(UJ) aBneB efuip uo |3Aai

«

£

lue

CQ

• c

c o

2

u

evi

"3 o

t u C

OU

a

1 ' m o T"~

'

= o : CO ; CO

: O = 1^-

r co o '

« K o§ >=5 -a r-C C O nj ^

13.« 3 ë

cT"

(~ (U

<« <

(poou

E o

art

! XL

rtou

m

sz V t™

(/) <t

ra

J_J

> ?

_ J

u} c i t n <N LX>

(uu) s6ne6 Efuir uo 13ASH

(JB9Â/ LU 01.) MOy

(Nia

Os J 3 00 >

'EVJ

195C

= o

p o

193

= o : CN

- o = a>

^ m >

O C3 0 0

'C ON

°7 .ao > £ u —<

J C

. * S o

, o i i

<o C3

11 = h- •+-

cti

Pi 3

S E ,

fa-i


exceptional in Khartoum in 1878, with prolonged periods of inundation in the town. The significance of high flows at Khartoum in the period 1869 to 1883 is, of course, that the estimated flows in the White Nile are lower than might have been expected from an analysis which relied solely on the post-1900 data for Khartoum.

Using the model from 1869, it is also possible to estimate the overall annual water balance of the Nile for four key locations: Aswan, the Blue Nile at Khartoum, the White Nile at Khartoum, and at the outlet of Lake Victoria (Fig. 2(c)). Where available, observed flows are shown, whilst in other periods the flows are estimated from the model. Once again, Khartoum flows are not shown from 1884 to 1895 since no flood flow data were available in this period. According to the model, in the period 1869-1883, in contrast to the Blue Nile flows, White Nile flows were high around 1878, but otherwise unremarkable. The estimated flows for Lake Victoria pre-1896 suggest a lesser, but sustained, increase in outflows in the 1890s, shortly before observations of levels first began.

These variations are also reflected in the estimated levels before 1896, which are shown in Fig. 2(d). The estimated mean annual level in 1878 is 13.32 m. This compares with peak levels of about 11.6m derived from a simple regression between Lake Victoria and Aswan flows by Flohn & Burkhardt (1985), and a value of about 14.2 m estimated by Nicholson (1998) from anecdotal evidence. Also, in the period 1870-1895, levels appear to have been slightly above the average in the period 1896-1960 (i.e. before the sudden rise in levels from 1961 to 1964) but, except for a brief period in the 1890s, were less than levels from 1964 to about 1990. Figure 2(d) also shows lines indicating the standard deviation of the estimated levels, which were typically about 0.25 m (giving a 95% confidence limit of about ±0.5 m). The symbols indicate estimates from anecdotal evidence which is described in the following section.

Comparisons with historical evidence

Although direct, enduring measurements of Lake Victoria levels did not begin until 1896, there are many accounts from government officials, travellers, residents and missionaries before that time which give an indication of lake levels, and these are summarized in Table 2 (see also Nicholson, 1998). This section reviews the main evidence available, both for Lake Victoria itself and locations downstream, and then compares these results with those from the model. The information is put into perspective by the fact that Speke first visited the Ripon Falls in 1862, establishing that a large river flowed northwards out of Lake Victoria, and that it was not until 1874 that Stanley established the extent of the lake.

Much of the key information is collected by Garstin (1901, 1904). In particular, Lyons contributed Appendix III to Garstin (1904) on the variations of level of Lake Victoria and reproduced the material in Lyons (1906).

Garstin (1901, p. 49) noted that "for many years previous to 1901, the mean levels of this lake have shown a steady fall, independent of the annual fluctuations due to the rainfall. This general lowering of the water-levels was noticed both by Sir H. Stanley and by Mason Bey at the time of their respective visits, while further proof is afforded by the register of the water-marks kept by the French Catholic Mission at Uganda. These last show that, three years ago [i.e. 1898], the mean water-level of the lake


averaged some 8 feet lower than it did twenty years earlier [i.e. 1878]." (our annotations in square parentheses). This statement was made on the authority of Mr McAllister, Vice-Consul at Wadelai and Sub-Commissioner of the Nile Province, who met the Garstin mission in 1901. Although this statement seems clear-cut, subsequent versions (e.g. Garstin, 1904; Ravenstein, 1901, who also misinterprets the date) place the mission variously at Mwanza, in Tanzania; at Bukumbi, on the south shore of the lake; and at Buganda, in Uganda. However, the key observation of an 8 feet drop (2.4 m) remains undisputed, except for a statement by Cdr Whitehouse (Garstin, 1904) that he could not see how such a drop could be possible "unless there was a similar rise between 1875 and 1877, which appears unlikely". His reservations were based on photographs of Ripon falls in 1875 (by Stanley) and 1900 which showed similar water levels at the falls.

Speaking of a visit in September 1889, Stanley (1890) himself states that the "French missionaries, since their settlement near the Lake at Bukumbi, have ascertained that the Lake is now three feet lower than when they first settled here—that is about eleven years ago—that Ukerewé is no longer an island but is a peninsula". Ukerewé island/peninsula is located in the southeast of the lake, and Stanley had noted in 1875 that the island was separated from the mainland by a narrow channel in places only three feet deep. Unfortunately, the French missionaries at Bukumbi "are said to have possessed a record of the level of the lake extending over many years, but it is reported to have been lost with much other scientific material by the sinking of a canoe on the lake." (Garstin, 1904).

The Appendix by Lyons in Garstin (1904) quotes several other descriptions of events in 1877 and 1878. For example, Wilson noted that in February 1877 there was exceptionally heavy rainfall to the south of the lake and an unusual rise of two feet. Hutchinson, quoting Wilson, states that the lake continued to rise and reached a

Table 2 Anecdotal and other evidence of Lake Victoria levels before 1896.

Date Level on Description available Jinja gauge (m)

End of year 1860 10.8 Peney's gauging March 1875 11.8 Stanley noted Ukerewe island channel 3 ft deep June 1875 12.0 Stanley poled through channel 3 ft+ deep February 1877 12.1 2 ft rise from mean May 1877 12.7 Further 2 ft rise 12 Jan. 1878 12.7 Same as previously observed 15 Mar. 1878 12.7 Same as previously observed Aug.-Sep. 1878 13.8 Levels 2.4 m higher than in 1898. Flooding in the Sudd swamps 1884 - Rains failed in northeast of lake Early 1886 - Rains failed in northeast of lake September 1889 11.0 Stanley noted Ukerewe no longer an island

11.8 French mission says lake level has fallen 3ft in the past 11 years April 1891 11.3 Level 5-6 ft below highest level in recent years

Rugezi straits impassable June 1892 13.1 Level 6 ft above normal January 1895 11.8 Lake rose 1.5 m in 1895 June 1895 13.3 Level higher than in 1892, less than in 1878. Flooding in the

Sudd swamps January 1896 11.57 First observed level at Entebbe (corrected to Jinja gauge) 1898 11.4 Based on Aug-Dec observations at Entebbe


maximum value in May 1877 of two feet above its February level. On his return to Kagehi (near Mwanza, on the southeast corner of Lake Victoria) on 12 January 1878, the level was within 1-1.5 inches of its May 1877 level, showing that the November-December rains had been particularly heavy. The level was the same on 15 March 1878, and a few days later in Uganda he received evidence of unusually high levels. Dr Felkin of the Royal Scottish Geographical Society, who visited Emin in Lado (near Mongalla) in 1878, recorded that the lake rose three feet above its normal level in August and September 1878 following exceptionally heavy rains.

Further evidence of later changes is given in the same Appendix by Lyons in Garstin (1904). Fischer noted that the rains failed to the northeast of the lake in early 1886, and also in 1884. This corresponds to Lyons' conclusion that 1880-1890 was a period of falling lake levels. In 1891, Gedge noted that levels were said to be about eight to nine feet below the highest water mark, but Dermott noted in April 1891 that south of Ukerewe the water level was five to six feet below the high level marks on the rocks; the Rugezi straits between Ukerewe and the mainland were not passable. In the following year, Lugard noted in June 1892 that following exceptionally heavy rainfall between November 1891 and February 1892 there had been a marked rise in the lake to "some 6 feet perhaps above its ordinary level". In 1892, Baumann also noted that the Rugezi straits were now passable and thus levels had risen since April 1891 following heavy rains. In 1895, Père Brard stated that the lake level on the southern coast had risen 1.5 m after heavy rains in March, and that plantations along the southern shore were destroyed; it was claimed that no such level had occurred for 30 years, but it was noted that 1878 was certainly as high or higher.

There are some inconsistencies in these observations, but an attempt has been made in Table 2 to interpret them in terms of Lake Victoria levels. There is also important evidence of an upper bound to lake levels in the recent past. An excavated cave at Hippo Bay near Entebbe (Brachi, 1960) revealed that beach sands overlain by occupation debris contained water-rolled fragments of charcoal distributed throughout the sands; a sample of charcoal was dated as 3720 ± 120 years BP (before present) (Bishop, 1969), and Bishop noted that "the recently recorded rise in the level of Lake Victoria between 1961 and 1964 is of interest as, at their maximum, the lake waters were within two feet of the dated beach gravel in Hippo Bay cave.. .This cave fill is so unconsolidated that even a few weeks would suffice to remove the whole deposit. It is certain that at no time since 3720 ± years ago has Lake Victoria risen to re-occupy the 10- to 12-foot cliff notch." An increase of 10-12 feet (3.05-3.66 m) over the Jinja gauge level for 21-31 August 1959 (10.80 m) gives a maximum of 13.85-14.46 m on the Jinja gauge as the highest level recorded in the past 4000 or so years.

Sudd inflows Narratives from the Sudd are also important as high inflows are an indication of high outflows from Lake Victoria. However, information on early measurements of Sudd inflows is limited. Garstin (1901, p. 49) refers to a measurement by Dr Peney on 27 February 1863 above Bedden rapids (above Rejaf), with a total of 550 m3 s"1, but Garstin (1904, p. 127) refers to 1861 and Lyons (1906, p. 91) gives a flow of 565 mJ s'1 on 27 February 1861. Although the accuracy of this measurement has been criticized, its significance is that, in February, at the end of the dry season, most of the flow downstream of Lake Albert originates from Lake Victoria, with inflows from the Semliki typically offset by losses in lakes Kyoga and


Albert, and the torrent flows being negligible. Thus the outflow at Jinja at the end of 1860 must have been of this magnitude, corresponding to a level of about 10.8 m on the Jinja gauge.

Within the Sudd, Johnson (1992) discusses large floods in 1878 and in 1895, damaging sorghum crops to the east of the Bahr el Zeraf. Luac and Ghol Dinka traditions recall that the flood arrived from the south, along the Bahr el Jebel, with little narrative evidence of unusually high flows on the Sobat. The flood of 1878 was dramatic, but did not produce any long-term displacement. A flood of 1895-1896 made an area to the east of the Bahr el Zeraf untenable.

Eighteen months of flooding began in October 1916, which submerged whole districts from Bor to Malakal and from Kongor to the Pibor River. The first flood of 1916-1917 came from the Bahr el Jebel, flooding Kongor in October. The flood of 1917 was higher and more prolonged, and was observed to come from two directions: from the main river and also from the east. This suggests that there was a large contribution from the Pibor, as the Sobat was high but the Baro at Gambeila was not high. There was also high rainfall around the Sudd in 1917-1918. This is some confirmation of the observation that the 1917-1918 event may have derived mainly from north of Lake Victoria, rather than the lake basin.

Floods of 1946-1947 are described by Johnson as due to heavy Sobat flows and subsequent "creeping flow", rather than runoff from the lake basin. The 1961-1964 flood was caused by heavy rainfall in 1961-1962 over a wide area. It can be noted that there was high flow on the Pibor in 1961-1962, which reached the Sobat in 1962-1963, before the peak outflow from Lake Victoria. The baseflow component at Malakal was higher and longer than in 1917.

There are few direct observations of Sudd areas. Those which do exist are mainly based on air photography (1930-1931), maps (from reconnaissance in 1950-1952) and satellite imagery (1979-1980). In order to estimate Sudd areas in those years for which no direct observations are available, water balance models are commonly used. The model used by Sutcliffe & Parks (1987) suggested that areas of the Sudd between 1905 and 1980 varied between approximately 5000 to 40 000 km2. In fact, the 1961-1964 event, when there was an approximate 20% increase in rainfall above average over Lake Victoria, caused a doubling of lake outflows, and caused the area of the Sudd to increase almost six-fold. Therefore, it is entirely reasonable to suggest that the high outflows from Lake Victoria in 1878 and 1895 must also have caused significant flooding.

Comparisons with the model results

Although there are contradictions in some of the early evidence of lake levels, Table 2 summarizes the results from these various descriptions and Fig. 2(d) shows the inferred values as points. Given that the model predicts annual mean values, on the graph the observations in Table 2 have been adjusted upwards by 0.25 m for dry season observations, and reduced by 0.25 m for wet season peaks, since the annual range (max to min) of Lake Victoria levels is typically about 0.5 m. The agreement between the model and historical observations is generally good, except for the 1895 event, when the anecdotal evidence suggests a peak about 0.9 m higher than given by


the model. However, given that the first observed level in January 1896 was only 11.57 m, perhaps the anecdotal value of 13.3 m in June 1895 is rather high, since that would imply a drop of 1.6 m in only six months.

For the main peak of 1878, the value of 13.3 m from the model agrees well with the inferred value of 13.55 m from the evidence of the French missionaries. The agreement in the general trend in levels is also good; for example, that levels rose suddenly in 1878, but dropped over the next 11 years, with rains failing in 1884 and 1886. Levels rose again in 1892 and 1895, before dropping back to the low levels recorded in the first few years of observation from 1896. The historical observations therefore add some credibility to the model results, and the implications for future flows of assuming this sequence of levels from 1869 are assessed in the following section.

DISCUSSION AND CONCLUSIONS

This study has provided new estimates of Lake Victoria levels and outflows for the period 1870-1895, extending the length of the observed data for Lake Victoria (1896 onwards) by some 25%. Crucially the analysis includes the major flood event of 1878, which anecdotal evidence suggests was the highest since 1860 (and possibly earlier); possibly linked to the El Nino event of 1878, which was one of the strongest of the past few centuries. This is supported by evidence from the Roda Nilometer, which indicates that the flows in 1878/79 were the highest for the previous fifty years or so, and therefore the highest in the nineteenth century.

Net basin supply

To assess the implications of these results for future levels and outflows at Lake Victoria, it is necessary to estimate the net basin supply to the lake, which is the sum of the inflows from rainfall and tributary flows, less the evaporation from the lake. The net basin supply is a true measure of climatic variability around the lake, whereas the lake level and outflow time series are not independent, and are influenced by storage effects in the lake, which can persist for the order of a decade.

A first estimate of the annual net basin supply can be obtained from the annual storage changes in the lake and the mean annual outflow. Observations suggest that the surface area of the lake only varies by a few percent about a mean value of 67 000 km2, so annual storage changes can be estimated by assuming a constant surface area equal to the mean. End-of-year levels have been estimated by assuming that the annual mean values calculated by the model are indicative of mid-year levels, and assuming a linear variation in levels between years. Although this is only a crude approximation, comparisons with the more sophisticated estimates of the annual net basin supply derived by Sene & Plinston (1994) showed good agreement over the period for which these estimates were available (1925-1990). Sene & Plinston (1994) also estimated the annual lake rainfall from raingauges around the lake, and Fig. 3(a) shows that there is a reasonable correlation between annual rainfall and net basin supply (expressed as a depth over the lake surface) over the period 1925-1990, with an uncertainty of about

316 Emma L. Tate et al,

±100 mm year" . The net basin supply estimates can therefore also be used to infer the lake rainfall, and Fig. 3(b) shows this comparison for the period 1870-1996.

Several interesting conclusions can be made from this plot. The first is that, during the 1878 event, both the rainfall and net basin supply rose rapidly to an exceptional level, but fell back to very low values in the next two years. This is in contrast to the

(b)

1000

-500 0 500 1000 Net basin supply (mm/year)

1500

3000

2500 -

2000 -

1500 -

1000 -

500

-500

Lake rainfall

-Net basin supply

1E70 1880 1890 1900 1910 1920 1930 1940 19501960 19701980 1990

Year

(c)

0.3

0.25

0.2

3 0.15

0.1

0.05

-1896-1996

-1870-1996

-400 -200 0 200 400 600 800 1000 1200 1400 1600

Net basin supply (mm/year)

Fig. 3 Results of modelling of Lake Victoria's net basin supply (a) correlation between annual rainfall and net basin supply; (b) net basin supply estimates used to infer rainfall over Lake Victoria; and (c) frequency distribution of the net basin supply.


1917 event, which built up over two to three years, and the 1961-1964 event, which arose initially from a rapid increase in rainfall, which was almost sustained for the next two to three years, and was followed by several years of above average rainfall. These three major events were therefore all qualitatively different, and the anecdotal evidence confirms this to some extent: i.e. the 1878 event arose from exceptional rainfall, affecting both the Blue and White Nile basins; the 1917 event arose from heavy rainfall to the north of the lake, with unusually high flows in the Pibor; whilst the 1961-1964 event arose from prolonged heavy rainfall in the Lake Victoria region, but not in the Blue Nile catchment. The smaller increase in levels during the 1890s appears to have been due to several years of slightly above average rainfall. It is also worth noting that, for all these events, the variations in areal rainfall are not particularly large, with a total range of only 1400-2400 mm year"1 over the period 1870-1998. However, due to the extreme sensitivity of levels to minor changes in rainfall, in terms of levels and outflows, these events appear much more dramatic than suggested by the rainfall data.

The statistical characteristics of the rainfall and net basin supply are summarized for several averaging periods in Table 3, using the standard WMO 30-year periods. In the period 1871-1900, the lake rainfall appears to have been slightly above average, and perhaps slightly more variable than usual. Only the years 1961-1990 stand out as having unusually high rainfall. The estimated maximum lake rainfall and lake levels reached are also shown; according to the model, the lake rainfall reached comparable peaks of 2300-2400 mm in 1878 and 1961, but a much lower peak in 1917. Annual mean levels in 1878 and 1961 were also similar. However, following the 1961 event, levels took much longer to decline due to the sustained increase in rainfall, and the lake levels more clearly exhibited the quasi-exponential decline expected from idealized models of the lake water balance (e.g. Sene & Plinston, 1994).

Table 3 Estimates for annual lake rainfall and net basin supply over various periods.

Period Lake rainfall (mm) Mean

Thirty year periods 1871-1900 1753 1901-1930 1707 1931-1960 1716 1961-1990 1837

Full record 1900-1996 1752 1871-1996 1754

St. dev.

190 145 176 179

171 175

Net basin (mm) Mean

407 319 302 598

404 407

supply

St. dev.

385 306 238 323

311 328

Max. rainfall period Value (mm)

2314 2015 2236 2370

2370 2370

in

Year

1878 1917 1951 1961

1961 1961

Max. annual level in period (m) Value (m)

13.3 11.7 11.5 13.0

13.0 13.3

Year

1878 1917 1942 1964

1964 1878

Possible future levels of Lake Victoria

In all stochastic models for net basin supply, the frequency distribution gives a useful first indication of the likely range of predicted levels, and Fig. 3(c) shows the estimated values for the periods 1896-1996 and 1870-1996 (i.e. excluding and including the results from the present study). In terms of future levels, a preliminary conclusion must be that these additional 26 years of data (1870-1895) do not greatly


change ideas about the natural flow regime of the lake, and stochastic projections of future flows are likely to be similar to previous estimates. Given the large financial investments in the Owen Falls Dam, and current plans to expand hydropower potential at sites further down the Nile, this is perhaps reassuring, although a much more detailed investigation would be required to draw conclusions which could be used in any economic assessments or detailed design studies. The additional 26 years of data therefore provide further evidence that Lake Victoria levels are highly variable, and can both rise and fall rapidly in periods of only a few years. Also, the 1961-1964 event, which at the time was a major surprise to those involved with development of the Nile, can be seen to be consistent with the natural rainfall regime of the region, with a return period which can be tentatively estimated as of the order 100 years.

Acknowledgements This paper was prepared with support from the UK Natural Environment Research Council (CEH Wallingford). The diagrams were designed and produced by Val Bronsdon and Sonja Folwell.

REFERENCES

Bishop, W. W. (1969) Pleistocene stratigraphy in Uganda. Memoir no. 10, Geol. Survey Uganda, Entebbe, Uganda. Brachi, R. M. (1960) Excavation of a rock shelter at Hippo Bay, Entebbe. Uganda J. 24, 62-70. Flohn, H. & Burkhardt, Th. (1985) Nile runoff at Aswan and Lake Victoria: a case of discontinuous climate time series.

Z. Gletscherkunde und Glazialgeologie 21, 125-130. Garstin, W. (1901) Report as to Irrigation Projects on the Upper Nile, etc. (In a Despatch from His Majesty's Agent and

Consul-General at Cairo). Parliamentary Accounts and Papers 91,1149 et seq. Garstin, W. (1904) Report upon the basin of the Upper Nile (In a Despatch from His Majesty's Agent and Consul-General

at Cairo). Parliamentary Accounts and Papers 111, 315-735. Hulme, M. (1990) The changing rainfall resources of Sudan. Trans. Instn British Geographers 15, 21-34. Hurst, H. E. (1952) The Nile. Constable, London. Hurst, H. E. & Phillips, P. (1933a) Ten-day mean and monthly mean gauge readings of the Nile and its tributaries. The

Nile Basin, vol. III. Government Press, Cairo, Egypt. Hurst, H. E. & Phillips, P. (1933b) Ten-day mean and monthly mean discharges of the Nile and its tributaries. The Nile

Basin, Vol. IV. Government Press, Cairo, Egypt. Johnson, D. H. (1992) Reconstructing a history of local floods in the Upper Nile Region of the Sudan. Int. J. African

Historical Studies 25(3), 607-649. Lyons, H. G. (1906) The Physiography of the River Nile and its Basin. Finance Ministry, Survey Department, Cairo, Egypt. Nicholson, S. E. (1998) Historical fluctuations of Lake Victoria and other lakes in the Northern Rift Valley of East Africa.

In: Environmental Change and Response in East African iMkes (ed. by J. T. Lehman). Kluwer, Dordrecht, The Netherlands.

Nicholson, S. E., Yin, X. & Ba, M. B. (2000a) On the feasibility of using a lake water balance model to infer rainfall: an example from Lake Victoria. Hydrol. Sci. J. 45(1), 75-95.

Nicholson, S. E., Some, B. & Kone, B. (2000b) An analysis of recent rainfall conditions in West Africa, including the rainy seasons of the 1997 El Nino and the 1998 La Nina years. J. Climate 13(14), 2628-2640.

Piper, B. S., Plinston, D. T. & Sutcliffe, J. V. (1986) The water balance of Lake Victoria. Hydrol. Sci. J. 31(1), 25-37. Ravenstein, E. G. (1901) The lake-level of Victoria Nyanza. Geogr. J. 18,403-406. Sene, K. J. (2000) Theoretical estimates for the influence of Lake Victoria on flows in the upper White Nile. Hydrol. Sci.

J. 45(1), 125-145. Sene, K. J. & Plinston, D. T. ( 1994) A review and update of the hydrology of Lake Victoria in East Africa. Hydrol. Sci. J.

39(1), 47-63. Stanley, H. M. (1890) In Darkest Africa, vol. II, 395. Sampson Low, London. Sutcliffe, J. V. & Parks, Y. P. (1987) Hydrological modelling of the Sudd and Jonglei Canal. Hydrol. Sci. J. 32(2), 143-159. Sutcliffe, J. V. & Parks, Y. P. (1999) The Hydrology of the Nile. IAHS Special Publ. no. 5. Walsh, R. P. D., Davies, H. R. J. & Musa, S. B. (1994) Flood frequency and impacts at Khartoum since the early

nineteenth century. Geogr. J. 160, 266-279. Yin, X. & Nicholson, S. E. (1998) The water balance of Lake Victoria. Hydrol. Sci. J. 43(5), 789-811.

Received 21 June 2000; accepted 13 November 2000

a water balance study of the upper white nile basin flows...

Documents