Hydrobiologia

, Volume 592, Issue 1, pp 151–173

Physical and chemical limnology of alpine lakes and pools in the Rwenzori Mountains (Uganda–DR Congo)

  • Hilde Eggermont
  • James M. Russell
  • Georg Schettler
  • Kay Van Damme
  • Ilse Bessems
  • Dirk Verschuren
Primary Research Paper

DOI: 10.1007/s10750-007-0741-3

Cite this article as:
Eggermont, H., Russell, J.M., Schettler, G. et al. Hydrobiologia (2007) 592: 151. doi:10.1007/s10750-007-0741-3

Abstract

This study describes the physical and chemical properties of 17 Afroalpine lakes (>2 m deep) and 11 pools (<2 m deep) in the Rwenzori mountains, Uganda-DR Congo, with the aim to establish the baseline conditions against which to evaluate future environmental and biological changes in these unique tropical ecosystems, and to provide the foundation for lake-based paleoenvironmental studies. Most Rwenzori lakes are located above 3,500 m elevation, and dilute (5–52 μS/cm specific conductance at 25°C) open systems with surface in- and outflow. Multivariate ordination and pairwise correlations between environmental variables mainly differentiate between (1) lakes located near or above 4,000 m (3,890–4,487 m), with at least some direct input of glacial meltwater and surrounded by rocky catchments or alpine vegetation; and (2) lakes located mostly below 4,000 m (2,990–4,054 m), remote from glaciers and surrounded by Ericaceous vegetation and/or bogs. The former group are mildly acidic to neutral clear-water lakes (surface pH: 5.80–7.82; Secchi depth: 120–280 cm) with often above-average dissolved ion concentrations (18–52 μS/cm). These lakes are (ultra-) oligotrophic to mesotrophic (TP: 3.1–12.4 μg/l; Chl-a: 0.3–10.9 μg/l) and phosphorus-limited (mass TN/TP: 22.9–81.4). The latter group are mildly to strongly acidic (pH: 4.30–6.69) waters stained by dissolved organic carbon (DOC: 6.8–13.6 mg/l) and more modest transparency (Secchi-disk depth: 60–132 cm). Ratios of particulate carbon, particulate nitrogen and chlorophyll a in these lakes indicate that organic matter in suspension is primarily derived from the lakes’ catchments rather than aquatic primary productivity. Since key features in the Rwenzori lakes’ abiotic environment are strongly tied to temperature and catchment hydrology, these Afroalpine lake ecosystems can be expected to respond sensitively to climate change and glacier melting.

Keywords

Alpine lakes Afro-alpine zone Climate change East African mountains Limnology Pools Rwenzori Water chemistry 

Introduction

The Rwenzori Mountains, the legendary ‘Mountains of the Moon’, straddle the equator along the border between Uganda and the Democratic Republic of Congo. Though surpassed in height by both Kilimanjaro (5,996 m) and Mt. Kenya (5,199 m), the high-mountain range as a whole is more extensive than either of these (Osmaston, 1989). Unlike the other mountains in equatorial East Africa, the Rwenzoris are not an extinct volcano, but comprise an uplifted complex of Precambrian rock (Ebinger, 1989). This complex has subsequently been dissected by erosion, resulting in six separate mountains all rising over 4,500 m: Mts. Stanley (5,109 m), Speke (4,889 m), Baker (4,842 m), Gessi (4,715 m), Emin (4,791 m) and Luigi di Savoia (4,626 m). Each of these consists of several peaks, the highest being Margharita on Mt. Stanley. All mountains were glaciated until historic times, but ice caps on Mts. Gessi, Emin and Luigi di Savoia have now completely disappeared. The Rwenzoris are dotted with numerous lakes mainly occupying glacially-excavated and moraine-dammed basins formed after the last glacial period (Osmaston, 1989; Osmaston & Kaser, 2001). Streams descending from the eastern slopes of the mountain range are generally small (Busulwa & Bailey, 2004), yet their sheer number discharges a large volume of water to the economically important Lakes Edward and George, and constitute the most important headwaters of the White Nile River north of Lake Victoria (Yeoman, 1989). As in other tropical high-mountain regions, the glaciers themselves are also important freshwater reservoirs that store precipitation during the wet season and release meltwaters during dry periods, thus buffering seasonal stream flow (Bradley et al., 2006). The Rwenzori therefore comprise a vital water catchment, upon which an estimated 500,000 Ugandans directly depend for their water supply (Howard, 1991). In addition, the mountain range holds an outstanding diversity of biota, with many species endemic to the Albertine Rift region (Kingdom, 1989). Recognising this importance, the Rwenzori were gazetted as a UNESCO World Heritage site in 1994.

Climate change and the associated retreat of Rwenzori’s glaciers may constitute an immediate threat to its mountain ecosystems. Observations of glacial termini confirm rapid glacial regression from 1906 to present. The retreat is consistent with warming of the tropical middle troposphere in recent decades (Hastenrath & Kruss, 1992; Gaffen et al., 2000; Taylor et al., 2006), though a longer-term decrease in precipitation and increased air humidity may also have affected glacier mass balance (Hastenrath, 2001; Kaser et al., 2004). At the current pace, all remaining glaciers will disappear within the next two decades (Kaser & Osmaston, 2002; Taylor et al., 2006). This in turn will directly affect hydrological processes and biology in the mountain lakes immediately downstream from these glaciers (e.g., Hauer et al., 1997; Koinig et al., 2002). Other factors that may increasingly affect these alpine lakes are the long-range atmospheric deposition of toxic pollutants, acids (e.g., Psenner & Schmidt, 1992; Mosello et al., 2001; Rogora et al., 2001) and dust (e.g., Psenner, 1999). Hence, establishing baseline data for the now still relatively pristine Rwenzori lakes will be critical to our understanding of the ecological and biodiversity effects of future climate and environmental changes in the region.

Previous scientific expeditions to the Rwenzori, primarily conducted during the early 20th century, focused almost exclusively on terrestrial biodiversity (e.g., Ogilvie-Grant, 1908; Heron, 1909; Oldfield, 1910; Burgeon, 1937; Hedberg, 1951; Kimmins, 1959; Salt, 1987; Wilson, 1995). Limnological surveys have been sparse and/or incomplete. De Heinzelin and colleagues conducted basic surveys of eight lakes on the Congo side of the Rwenzori range (de Heinzelin & Mollaret, 1956; Mollaret, 1961). Subsequent studies focused only on four lakes on the Ugandan side, located along tourist routes in the Bujuku-Mubuku river drainage (Löffler, 1964, 1968a, b; Livingstone, 1962, 1967). Cholnoky (1964) visited seven lakes in the remote Kamusongi and Nyamughasana valleys to inventory their lacustrine diatom flora, but did not collect morphological, physical or chemical data on these lakes.

This study aims to create a reference database on the physical and hydrochemical characteristics of a representative number of high-elevation lakes and pools in the Rwenzori Mountains, to permit monitoring of how climate-driven environmental change will affect their unique fauna and flora, and their ecosystem functioning. As on Mt. Kenya (Löffler, 1964, 1968a, b) and mountain ranges elsewhere (e.g., Marchetto et al., 1994; Kamenik et al., 2001), we expect the physical and chemical properties of Rwenzori lakes to be strongly linked to elevation (mean annual temperature), local catchment characteristics (type of vegetation cover, bedrock composition), and whether or not a lake is fed directly by glacial meltwater. If this is indeed the case, Rwenzori lakes and their aquatic biota will likely undergo major impacts from current climate warming and predicted disappearance of the glaciers. In view of conserving a potentially unique tropical cold-stenothermic aquatic fauna, we also aim to appraise whether deepwater benthic environments in thermally-stratified lakes at lower elevations could substitute for shallow-water benthic environments in (un-)stratified lakes higher up the mountain. Finally, the assembled data on how key environmental variables (bottom temperature and oxygen, pH, transparency and nutrients) are structured among lakes within the study region will support studies relating the distribution of aquatic biota to local environmental gradients, and the use of their remains preserved in lake-sediment records as proxy indicators in paleoenvironmental reconstruction.

Materials and methods

Description of the study sites

All but one of the lakes surveyed for this study are located on the Ugandan side of the mountain range between 0°16′–0°24′ N and 29°52–29°59′ E, and part of the Batoda, Bujuku, Butawu, Kamusongi, Mahoma and Nyamugasani River drainages (Fig. 1; Table 1). The exception is Lac du Speke, headwater lake of the Luusilubi River and situated immediately across the Ugandan border in DR Congo. With few exceptions the Rwenzori lakes were formed by glacial activity (Table 1). During the Last Glacial Maximum (LGM, 21 kyears BP) the local snowline extended down to 3,000 m (Mahaney, 1989), but since most lakes are located above 3,700 m they are probably of Holocene age; at least some of their basins were formed only following glacier retreat after the Little Ice Age (de Heinzelin, 1962). Most lakes (e.g., Batoda, Bigata, and Kachope) were created after a glacial valley was dammed by terminal or recessional moraines, but those above 4,200 m occupy glacially-scoured basins. Some of the latter are quite young (i.e., ∼60 years or less; Osmaston, 2006), as their basin was exposed by recent glacier regression. Examples include Lake Ruhandika at the foot of Speke glacier, and Lake Irene below Stanley glacier. All rock pools above 4,400 m are also formed by glacial scouring below the headwall of former glaciers; those at lower elevations are marsh or river features. Lake Bujuku appears to be dammed by a landslide off the North slope of Mt. Baker rather than a moraine (Livingstone, 1967). Lake Mahoma is located at 2,990 m within the LGM terminal moraine, and is extraordinary in that its size, shape, and depth all suggest that unlike any other lake in the African tropics (Livingstone, 1967) its basin was formed after the thawing of a block of ice detached from the retreating glacier (i.e., a kettle lake; Wetzel, 1983). In many lakes, shoreline features such as wave cuts in unconsolidated material and water marks on rocks and large boulders showed evidence of seasonal lake-level changes, presumably linked to river discharge variation between wet and drier seasons. However, all of the lakes are hydrologically open, limiting the extent to which their water level can fluctuate.
Fig. 1

Map of the Rwenzori Mountains, showing glaciers, river drainages and location of the 17 study lakes (numbers refer to their listing in Table 1)

Table 1

Overview of the study sites (17 lakes and 11 pools) listed per drainage, with indication of their origin, mixing regime (see text), and major environmental variables

No.

Site name

Sample date

LAT (N)

LONG (E)

Location

Catchment type

Origin

Mixing regime

ELEV (m asl)

Zmax (m)

Lmax (m)

Area (ha)

TEMPbottom (°C)

CONDbottom (μS/cm)

O2 surface (mg/l)

O2 bottom (mg/l)

pHbottom (units)

SECCHI (cm)

Lakes

1

Lac du Speke

7/15/2006

0°24.321′

29°52.869′

Mt. Speke

Alpine

Glacially-scoured

2a-b

4,235

17.0

275

3.912

5.15

12

7.96

5.04

5.20

120

2

Mahoma

7/27/2006

0°20.734′

29°58.102′

Mahoma

Forest/Bamboo

Kettle

3

2,990

25.6

325

4.763

12.52

41

5.36

0.45

5.10

82

3

Bujuku

7/11/2006

0°22.688′

29°53.576′

Mubuku-Bujuku

Alpine*

Landslide-dammed

2b

3,891

13.5

400

7.873

6.95

50

7.40

0.24

5.88

180

4

Ruhandika

7/12/2006

0°23.453′

29°53.433

Mt. Speke

Nival

Glacially-scoured

N/A

4,341

3.0

30

0.009

(4.00)

5

7.59

4.37

6.03

(166)

5

Irene

7/14/2006

0°22.936′

29°52.857′

Mt. Stanley

Nival

Glacially-scoured

N/A

4,487

3.5

50

0.016

(4.00)

7

7.59

4.37

5.70

(186)

6

Upper Kitandara

7/19/2005

0°21.171′

29°53.241′

Butawu

Alpine

Moraine-dammed

2a

4,018

14.5

300

3.258

4.91

28

9.40

7.39

7.00

280

  

7/17/2006

          

4.72

32

7.86

7.82

6.75

280

             

4.82

30

8.63

7.61

6.88

280

7

Lower Kitandara

7/17/2006

0°20.947′

29°53.194′

Butawu

Alpine

Moraine-dammed

2a

3,989

11.0

300

2.860

5.93

33

7.47

7.09

6.45

240

8

Lower Kachope

17–18/07/2005

0°20.066′

29°52.296′

Butawu

Ericaceous

Moraine-dammed

2a-b

3,841

8.3

200

0.812

7.35

22

7.66

5.75

6.29

130

  

7/19/2006

          

7.43

25

6.46

4.61

5.76

110

             

7.39

23.5

7.06

5.18

6.03

120

9

Middle Kachope

17–18/07/2005

0°20.040′

29°52.418′

Butawu

Ericaceous*

Moraine-dammed

2a

3,843

3.3

150

1.007

7.33

27

7.60

6.57

6.56

125

10

Upper Kachope

7/16/2005

0°19.932′

29°52.567′

Butawu

Ericaceous

Moraine-dammed

2a

3,961

12.0

350

4.593

6.36

19

8.85

6.78

6.86

120

  

7/19/2006

          

6.36

22

6.53

5.10

5.53

110

             

6.36

20.5

7.69

5.94

6.20

115

11

Batoda

7/14/2005

0°17.977′

29°52.972′

Batoda

Ericaceous

Moraine-dammed

2a

4,017

15.0

450

9.530

6.09

12

7.53

5.95

5.08

80

  

7/21/2006

       

15.5

  

6.10

12

6.79

6.30

4.84

90

             

6.10

12

7.16

6.13

4.96

85

12

Kopello

7/12/2005

0°18.612′

29°53.504′

Nyamugasani

Alpine*

Moraine-dammed

2b

4,017

14.5

350

3.007

5.51

10

9.55

2.31

4.88

112

13

Bigata

7/11/2005

0°18.396′

29°53.540′

Nyamugasani

Alpine*

Moraine-dammed

3

3,983

18.0

200

1.441

5.17

24

8.00

0.22

5.74

132

14

Africa

7/9/2005

0°17.697′

29°53.770′

Nyamugasani

Alpine*

Moraine-dammed

2b

3,943

4.7

270

1.001

5.48

19

9.39

3.30

5.37

90

15

Kanganyika

7/8/2005

0°17.189′

29°53.847′

Nyamugasani

Ericaceous

Moraine-dammed

2a

3,822

25.0

750

11.234

6.95

13

7.82

5.11

5.12

85

  

7/23/2005

          

7.31

14

6.17

5.87

4.73

100

             

7.13

13.5

7.00

5.29

4.93

92.5

16

Katunda

7/23/2006

0°16.798′

29°53.728′

Nyamugasani

Ericaceous

Moraine-dammed

1

3,782

10.5

400

3.765

7.52

14

6.35

6.44

5.22

117

17

Nsuranja

7/6/2005

0°17.579′

29°54.501′

Nyamwamba

Ericaceous*

Moraine-dammed

2a-b

3,834

12.5

300

2.155

6.78

12

7.17

3.80

4.31

60

Pools

18

Kamusongi’s pool

7/20/2005

0°22.340′

29°53.003′

Mt. Stanley

Nival

Glacially-scoured

N/A

4,509

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

19

Mbahimba’s pool

7/20/2005

0°22.371′

29°52.909′

Mt. Stanley

Nival

Glacially-scoured

N/A

4,555

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

20

Muhesi’s pool

7/20/2005

0°22.367′

29°52.888′

Mt. Stanley

Nival

Glacially-scoured

N/A

4,570

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

21

Balengekania’s pool

7/14/2006

0°22.979′

29°52.739′

Mt. Stanley

Nival

Glacially-scoured

N/A

4,573

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

22

Salomon’s pool

7/14/2006

0°22.946′

29°52.859′

Mt. Stanley

Nival

Glacially-scoured

N/A

4,480

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

23

Zaphana’s pool

7/14/2006

0°23.004′

29°52.773′

Mt. Stanley

Nival

Glacially-scoured

N/A

4,543

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

24

Tuna Noodle pool

7/15/2005

0°20.059′

29°52.353′

Bamwanjara pass

Alpine*

Marsh pool

N/A

4,380

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

25

Josephat’s pool

7/19/2005

0°21.676′

29°53.252′

Butawu

Alpine

Marsh pool

N/A

4,100

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

26

Baguma’s pool

7/7/2005

0°17.843′

29°53.986′

Nyamugasani

Ericaceous*

Marsh pool

N/A

3,994

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

27

Zephania’s pool

7/13/2005

0°18.385′

29°53.083′

Nyamugasani

Alpine*

Marsh pool

N/A

4,224

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

28

Mutinda pool

7/5/2005

0°16.465′

29°55.651′

Kamusongi

Ericaceous*

River pool

N/A

3,507

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

mean (n = 7)

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

minimum

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

maximum

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

SD

<2

<10

<0.003

N/A

N/A

N/A

N/A

N/A

N/A

Abbreviations are as defined in the text; N/A = no data available. Values in bold typeface are the means of two measurements in consecutive years; T and SECCHI values for Irene and Ruhandika (between brackets) are not measured but estimated as explained in the text

The Rwenzori are wetter than other East African mountains, with annual rainfall varying with altitude from 2,000 to 3,000 mm, and being heaviest on the eastern slope, which faces the prevailing winds. On the Uganda side heavy rain can occur any time of year, but the most rainy periods are from mid-March to May and from September to mid-December (Temple, 1961; Osmaston, 1965). The equatorial position of the mountain range creates daily air temperature oscillations between − 5 and 20°C in the Alpine and Nival zones, an order of magnitude greater than the seasonal variation in maximum daytime temperature. Occasional night-time freezing occurs from ∼3,000 m altitude (the present-day boundary between Bamboo and Ericaceous zones); at 4,000 m (the Ericaceous-Alpine zone boundary) freezing occurs on 80–90% of the nights (Rundel, 1994).

The Rwenzori, like other tropical mountains, exhibit discrete vegetation belts distributed across the altitudinal gradient (Fries & Fries, 1948; Hedberg, 1951). Except for Lake Mahoma, which is surrounded by a mix of montane and bamboo forest, all study sites are located in the Ericaceous (∼3,100–3,800 m), Alpine (∼3,800–4,300 m) or Nival zones (>4,300 m). The Ericaceous zone is characterized by broad-leaved trees (Hypericum spp., Hagenia abyssinica and Rapanea rhododendroides), arborescent heathers (Erica spp.) draped with lichens (Usnea spp.), flowering shrubs (typically Helicrysum guilelmii), scattered tree groundsel (mostly Senecio longeligulatus), and giant lobelias (Lobelia stuhlmannii). The most abundant plants in the Alpine zone are Carex tussocks, as well as Helichrysum stuhlmannii and Alchemilla spp. Various Senecio and Lobelia spp. (primarily S. adnivalis, S. friesiorum and L. wollastonii) are common in ravines and other sheltered or well-watered sites. The Nival zone is typified by bare rock, moss, liverworts and lichens. The altitudinal vegetation belts are not always clearly separated, and transition zones with mixed vegetation are common throughout the range.

Sample collection and laboratory analyses

In July of 2005 and 2006, 17 lakes (depth Zmax > 2 m; surface area at least 90 m2 but usually >1 ha) and 11 pools (Zmax < 2 m; surface area < 30 m2) were sampled (Table 1). Because of their small size, pools lack an official name; we named them after guides, porters, or people from the research team. Latitude (LAT), longitude (LONG) and elevation (ELEV, in meter above sea level) were recorded by GPS (Garmin eTrex Vista; mean error 10 m). Water samples were taken ∼30 cm beneath the surface at mid-lake locations, but near the shore in pools. Lake bathymetry was determined by GPS-guided echosounding (handheld depth sounder Echotest II), and bathymetric maps were drawn in Surfer v. 7.0 (Golden Software, Inc.) using kriging as the gridding method. Lake surface area (Area) and maximum length (Lmax) were subsequently calculated using ImageJ v. 1.37 software (W. Rasband, unpublished software); for pools these morphometric parameters were visually estimated from the shore.

Temperature (T), pH, oxygen (O2) and conductivity (COND: specific conductance at 25°C, in μS/cm) of the water column were measured at 1-m intervals with a Hydrolab Quanta CTD profiler. Transparency (SECCHI) was measured using a 22-cm diameter Secchi disk. Due to logistic constraints no T or O2 measurements were done in the pools, and their pH and conductivity was measured up to 4 h after sampling. The majority of lakes and pools were sampled only once, but repeat sampling of five lakes in 2005 and 2006 indicates quasi-stationary dry-season conditions in consecutive years. Samples for analysis of cations (Ca, Mg, Na, K, Li, Ba, Sr, Fe, Mn), anions (F, Cl, NO3, SO4), dissolved silica (Si), total dissolved sulfur (TS), dissolved phosphorus (PO4-P), dissolved inorganic carbon (DIC) and dissolved organic carbon (DOC) were filtered through Sartorius\(^{\circledR}\) cellulose acetate filters of 0.45 μm mesh. Filtrates for cation, PO4-P and TS analysis were stabilized with nitric acid (Merck, Suprapur quality) and stored in 50 ml pre-cleaned polyethylene bottles. Filtered samples for analysis of anions, Si and dissolved carbon were stored untreated in separate bottles. For determination of total phosphorus (TP) and total nitrogen (TN), unfiltered samples were fixed by adding concentrated sulfuric acid. Samples for analysis of particulate organic carbon (POC), particulate organic nitrogen (POC) and chlorophyll a (Chl-a) were taken by filtering 180–420 ml water through pre-ashed Whatman\(^{\circledR}\) glass microfibre filters (0.45 μm). The filters carrying suspended material were folded in half, placed in plastic Petri dishes and wrapped in aluminium foil. All water samples were kept dark and cool (filters were stored frozen) during the field campaign and until their shipment to the laboratory.

Cations, Si, PO4-P and TS were analysed by Inductively Coupled Plasma Atom Emission Spectrometry (ICP-AES; IRIS, Thermo Elemental). F, Cl, NO3 and SO4 were analysed by ion-exchange chromatography (DX100, Dionex). The highly dilute nature of the studied waters (5–52 μS/cm specific conductance at 25°C) resulted in measured values for several chemical species (K, Na, PO4-P, F, Cl, SO4, NO3) that are close to or below the detection limit of the analytical methods employed. Risk of contamination was minimized by repeated pre-flushing of membrane filters and autosampler vials with the individual waters. We also took account of the non-linearity of IC calibration curves by increasing the density of calibration points in the low-concentration range. Summation of all the measured ion concentrations yielded a systematic excess of calculated conductivity relative to conductivity measured in the field. We tentatively attribute this to the leaching of cations from microscopic (<0.45 μm) organic particles remaining in the filtered water samples, after field addition of nitric acid. However, no systematic variation in cation excess occurs in relation to the time elapsed between sample collection and analysis.

DIC and DOC (excluding volatile organic carbon components, i.e. the Non-Purgeable Organic Carbon or NPOC) were measured by IR-spectrometry (TOC-2000A, Shimadzu). The NPOC measurement protocol included acidification of the sample by addition of hydrochloric acid, sparging of the sample with high purity air, and catalytic burning of the sample aliquots at 680°C. DIC and field pH data were combined to calculate the in situ HCO3, CO3 and CO2 concentrations, following Wetzel & Likens (1990). TP was determined by wet oxidation in an acid persulphate solution (120°C, 30 min), and TN (as nitrate plus nitrite) by wet oxidation in an alkaline persulphate solution (120°C, 30 min), both following Grasshoff et al. (1983). Chl-a was determined by high-pressure liquid chromatography (HPLC), following the protocol of Wright et al. (1991, 1997). POC and PON were measured by flash combustion/thermal conductivity in a CE Instruments NC2100 elemental analyzer. The reported POC and PON values are means of duplicate measurements, using subsamples of each filter.

Data screening and statistical analyses

The substantial differences in size and depth ranges of Rwenzori lakes and pools (Table 1) justifies separate analysis of their thermal and chemical regimes. This study focuses on the comparative limnology of the true lakes, with pools referred to only in passing. The initial data set thus includes 17 sites and 39 environmental variables (see Electronic supplementary material). In lakes sampled twice, we use the average values of the two measurement series. Four variables were removed because their concentrations were below detection limit at more than half of the study sites (NO3, PO4-P, Li) or were calculated as negligible (CO3, due to low pH). The remaining variables were tested for normality using Shapiro–Wilk tests (Shapiro et al., 1968). Eight variables were normalized using either log (Tsurface, Tbottom, CONDsurface) or log (x + 1) transformation (SO4, TS, TPU, H2CO3, HCO3). Two variables (F and Mn) had to be excluded because their distribution could not be normalized.

We calculated a Pearson correlation matrix to quantify relationships between the 33 environmental variables retained for multivariate analysis. We then used principal components analysis (PCA) to identify the principal environmental gradients structuring the data set. Some variables which are intrinsically strongly correlated with others were excluded to avoid distortion of the analysis due to redundancy. Specifically, Lmax was removed in favour of lake Area. Surface-water measurements of T, COND and pH were removed in favour of the corresponding bottom water measurements, because the latter reflect long-term conditions while surface-water data are more subject to daily (i.e. depending on the time of day) and seasonal fluctuations (i.e. whether the measurement occurred just before or after a deep-mixing event). Surface and bottom water measurements of oxygen, on the other hand, were both kept in because lack of correlation between the two suggests that site-specific basin morphometry or primary production affects local bottom oxygen regimes. Missing Tsurface and SECCHI data for Lakes Irene and Ruhandika were estimated using the significant regressions between Tsurface and ELEV and between SECCHI and DOC at the other 15 sites. The Tbottom for these two lakes was set at 4°C, because (1) the regression between Tbottom and ELEV in the other lakes indicated that their Tbottom is unlikely to be significantly higher than 4°C; and (2) estimated Tsurface suggests at least daytime warming of their water surface.

For PCA analysis we further added a set of categorical variables representing the dominant vegetation type in each lake catchment, namely: bare rocks (Nival), alpine vegetation (Alpine), alpine vegetation dominated by Carex swamp (Alpine*), Ericaceous vegetation (Ericaceous), Ericaceous vegetation dominated by Carex swamp (Ericaceous*), and a mix of montane and bamboo forest (Forest/Bamboo). The final set of 35 variables was centred and standardized to allow comparison of disparate variables (ter Braak & Šmilauer, 1998). We refrained from testing the significance of the PCA ordination axes since the theory that has been developed for these tests suffers intrinsic methodological flaws (ter Braak & Šmilauer, 2002). Finally, we performed redundancy analysis (RDA) to assess whether selected external variables significantly influenced the physical and chemical properties of Rwenzori lakes. For this we used six predictor variables (ELEV; Ericaceous; Ericaceous* + Alpine*, Alpine, Nival, and Forest/Bamboo) and 26 response variables (the same set as for the PCA, but excluding the predictor variables and morphometric parameters), and performed a series of RDAs constrained to one single predictor at a time. Statistical testing of every predictor variable was done with the random permutation procedure of CANOCO (reduced model, 999 permutations; ter Braak & Šmilauer, 2002) with manual variable selection.

Ternary graphs, correlation matrices and normality tests were generated using the software package STATISTICA 5.5 (Statsoft, 2000). Multivariate statistics were performed using the package CANOCO v. 4.5 (ter Braak & Šmilauer, 2002).

Results

Multivariate statistical analysis

The first two PCA axes together account for 51.9% of the environmental variance in our Rwenzori lake data set (λ1 = 0.298 and λ2 = 0.221; Fig. 2). PCA axis 1, which explains 29.8% of the total variation, mainly captures gradients in land cover (e.g. Nival: presence/absence of bare rocks), elevation, bottom water temperature, major cations, dissolved Si, and levels of the trace elements Ba, Fe and Sr. Bottom water temperature and elevation point in opposite directions, reflecting the strong inverse relationship between them (see correlation matrix, see Electronic supplementary material). All lakes located below 4,000 m plot in the left quadrants, and with the exception of Batoda all lakes above 4,000 m plot in the right quadrants. Lake Africa and the Kachope lakes have Mg, Na, K, Sr and Si concentrations above the mean, explaining their clustering with these variables in (or close to) the upper left quadrant. Lakes in the lower left quadrant (e.g. Nsuranja, Batoda, Kanganyika) are typified by enhanced Ba and Fe concentrations. PCA axis 2, which explains 22.1% of the total variation, predominantly captures gradients of Secchi depth, DOC, TN, TP, major anions, and bottom pH. Here, Secchi depth and DOC point in opposite directions, reflecting the strong inverse relationship between them. As reflected by the narrow angles between variables pointing towards the lower left quadrant, less transparant lakes (high DOC content) are more acidic and hold more nutrients (high TP and TN); they also tend to be larger and deeper than other lakes. The opposite is true for lakes clustering in (or close to) the upper right quadrant. As indicated by the land-cover centroids, the former type of lakes are predominantly surrounded by Forest/Bamboo, Ericaceous vegetation and/or Carex swamp, whereas the latter type are enclosed by bare rocks or Alpine vegetation. Some lakes from the same valley (drainage) but in different vegetation zones plot in the same quadrant, e.g. Africa, Kanganyika and Katoda.
Fig. 2

Principal Components Analysis (PCA) biplot showing relationships between the 17 Rwenzori study lakes (full circles) and 35 environmental variables (arrows and open circles)

RDA analysis indicates that elevation (F = 3.400, P = 0.003) and Nival (i.e. presence/absence of bare rocks; F = 3.915, P = 0.007), Alpine (F = 2.297; P = 0.026) and Ericaceous* + Alpine* vegetation (i.e. presence/absence of Carex swamps; F = 2.376, P = 0.035) contribute significantly and independently to explained variation in within-lake environmental conditions: RDA axes 1 and 2 together explain 41.1% of the observed variance (λ1 = 0.277 and λ2 = 0.134).

Lake morphometry and catchment characteristics

Lake morphometry is summarized in Table 1 and illustrated in Figs. 35. Zmax has a significant (P < 0.05) relationship with both Lmax (r = 0.57) and Area (r = 0.60). Lake Kanganyika is by far the largest and deepest of all lakes (Zmax = 25.0 m; Lmax = 750 m; Area = 11.2 ha), while Lake Ruhandika is the smallest and shallowest (Zmax = 3.0 m; Lmax = 30 m; Area = 0.009 ha). Lakes dammed by moraines or land-slides in steep valleys (e.g. the Kachope and Kitandara lakes; Figs. 3, 4) are typically long-drawn between in- and outflow, whereas lakes in glacially-scoured basins (e.g. Lac du Speke; Fig. 3) and the Mahoma kettle (Fig. 3) are more circular and lack a well-defined inlet. Most moraine-dammed lakes have a single major inflow (e.g. Upper and Lower Kitandara; Fig. 3), in some it is supplemented by several smaller inflowing streams that drain the local catchment (e.g. Batoda and Bujuku; Figs. 3, 4). None of the lakes is steep-sided all around, so they all have (at least locally) a clear littoral, shallow-water zone. Except for lakes Ruhandika and Irene in the Nival zone whose basin lack continuous vegetation cover, all surveyed Rwenzori lakes are surrounded either by bogs with Carex and Sphagnum (e.g. Africa, Bigata, Middle Kachope, Nsuranja), or steep rocky slopes with Ericaceous (e.g., Kanganyika, Katunda, Upper and Lower Kachope) or Alpine vegetation (e.g., Bujuku, Kopello, Kitandara, Speke), as defined above. This vegetation is still pristine in all drainage basins except the Kachope lakes (Fig. 4), where deliberately-set recent fires burned much of the vegetation.
Fig. 3

Bathymetry and surrounding land cover of Upper and Lower Kitandara lakes (Butawu drainage), Lac du Speke (Mt. Speke), Mahoma (Mahoma drainage) and Bujuku (Mubuku-Bujuku drainage)

Fig. 4

Bathymetry and surrounding land cover of lakes Nsuranja (Nyamwamba drainage), Upper, Middle and Lower Kachope (Butawu drainage), and Batoda (Batoda drainage)

Fig. 5

Bathymetry and surrounding land cover of lakes Kopello, Afrika, Bigata, Kanganyika and Katunda (Nyamugasani drainage)

Thermal regime

None of the investigated Rwenzori lakes had ice cover during our visits. Excluding the atypically warm Lake Mahoma at 2,990 m (13.5°C), mid-day surface temperatures ranged from 5.5°C in Irene to 9.1°C in Bigata, with a mean of 7.8°C (Table 1). Bottom water temperatures ranged from 4.8°C in Upper Kitandara to 7.5°C in Katunda, again excluding Mahoma (12.5°C). The expected inverse correlation between water temperature and elevation is distinctly stronger using values of bottom temperature (r = − 0.93, P < 0.001) than of surface temperature (r = − 0.82; P < 0.001). When excluding Lake Mahoma, the relationship of surface temperature with elevation even lacks statistical significance (r = − 0.44, P > 0.05; versus r = − 0.84, P < 0.001 for bottom temperature).

Water-column profiles of temperature, dissolved O2, conductivity and pH (Fig. 6) indicate that all surveyed Rwenzori lakes undergo stratification, also the shallow lakes Africa (Zmax = 4.7 m; Fig. 6d) and Middle Kachope (3.3 m). Based on the magnitude of surface-to-bottom contrast in the above-mentioned parameters observed at the time of our visit(s), we differentiate three types of mixing regime (Table 1). Type 1 is defined for Lake Katunda (Fig. 6a), the only lake where temperature, oxygen, pH and conductivity were uniform at all depths, except for modest mid-day warming in the upper 1 m. As normal wind-driven mixing in this lake most probably extends deeper than 1 m, we surmise that the lack of a clear thermocline in our profile reflects an event of deep convective mixing shortly before the measurement was taken. The type 2 mixing regime comprises lakes with a slightly to strongly clinograde oxygen curve, but where the bottom water remains well-oxygenated ([O2] > 1 mg/l). These lakes display a more or less well-developed thermocline at 3–5 m depth and (often) reduced pH values in the hypolimnion, but no significant conductivity gradient between surface and deep water. In the context of benthic habitat conditions we further differentiate between lakes with hypolimnetic O2 concentrations close to those at the surface (Type 2a; e.g., Kanganyika, Fig. 6b) and lakes where hypolimnetic O2 has been markedly depleted (Type 2b; e.g., Kopello and Africa, Fig. 6c, d). Type 3 comprises lakes with a strongly clinograde oxygen curve, and anoxic bottom waters with significantly higher conductivity than surface waters (Fig. 6e, d).
Fig. 6

Profiles of temperature, oxygen, conductivity and pH in lakes Katunda (a), Kanganyika (b), Kopello (c), Africa (d), Bigata (e) and Mahoma (f). In case of repeated measurements, full and dotted lines indicate profiles of 2005 and 2006, respectively

Transparency, water chemistry and primary production

Secchi depth transparency in the 17 study lakes ranged from 60 to 280 cm (Table 1), and DOC ranged from 3 to 13.6 mg/l (see Electronic supplementary material). F. Secchi depth was significantly correlated with both DOC (r = − 0.66; P < 0.01) and Fe (r = − 0.61; P < 0.05), which also co-vary among them (r = 0.79; P < 0.001). Low-transparency lakes were either tea-coloured (e.g. Nsuranja, Batoda, Africa, Kanganyika) or green (Mahoma only).

Surface-water conductivity ranged from 5 to 52 μS/cm (mean 17.3 μS/cm) in the lakes, and from 5 to 18 μS/cm (mean 9.6 μS/cm) in the pools (Table 1). The principal cation concentration sequence in the lakes is Ca > K > Na > Mg, with means of 1.953, 0.511, 0.420 and 0.346 mg/l, respectively (see Electronic supplementary material for individual values). Pools display a similar cation concentration pattern (Fig. 7a), with means of 0.867, 0.548, 0.359 and 0.226 mg/l. Anion concentration patterns exhibit considerably less coherence (Fig. 7b), with the sequence SO4 > HCO3 > Cl in lakes of the Bujuku and Butawu drainages (means of 1.608, 0.672 and 0.414 mg/l, respectively) and SO4 > Cl > HCO3 in lakes on Mt. Speke, Mt. Stanley (including the rock pools) and in the Mahoma, Batoda and Nyamughasani drainages.
Fig. 7

Ternary diagram showing major cation (a) and anion (b) proportions of individual Rwenzori lakes (full circles) and pools (open circles), with codes as in Table 1. Also indicated (as square symbols) are the mean values for the Ugandan Rwenzori lakes (R1, n = 17; this study), Rwenzori pools (R3, n = 7; this study), Congolese Rwenzori lakes (R2, n = 5, derived from de Heinzelin & Mollaret, 1956), and selected freshwater (<1,000 μS/cm) lakes in the Ugandan lowlands (L1, n = 13, derived from Kilham 1971; L2, with n = 9, derived from Kizito et al., 1993; L3, with n = 21, Verschuren et al., unpublished data). Kizito et al. (1993) and de Heinzelin & Mollaret (1956) do not report anion data

Surface-water pH in the lakes ranged from 4.30 to 7.80, and is significantly (P < 0.01) correlated with DOC (r = − 0.68), Ca (r = 0.78) and DIC (r = 0.71). Dissolved Si concentrations in the surface water varied from 0.11 to 1.90 mg/l but with a fairly high mean value of 1.21 mg/l: in all but four lakes, surface Si concentrations were near or above 1 mg/l. The four exceptions are the two lakes in the Nival zone (Ruhandika: 0.11 mg/l; Irene, 0.14 mg/l), Lac du Speke on the DR Congo side of the range (0.26 mg/l), and our July 2006 measurement in Upper Kachope (0.22 mg/l; see Electronic supplementary material).

Surface-water TP concentrations in the surveyed lakes ranged from 3.1 to 68.4 μg/l. TN ranged from 155.4 to 695.8 μg/l, and was significantly (P < 0.01) correlated with DOC (r = 0.76). TN:TP varied between 9:1 and 81:1, Chl-a between 0.29 and 10.9 mg/l, POC:Chl-a between 85:1 and 2,471:1, and POC:PON between 11:1 and 14:1 (see Electronic supplementary material).

In agreement with the ordination of lakes and environmental variables in PCA (Fig. 2), distinct patterns among Rwenzori lakes with regard to the inter-related variables of Secchi depth, DOC, pH, TP, TN:TP and Chl-a permit the distinction of two groups of lakes. The first group includes six lakes (Lac du Speke, Bujuku, Ruhandika, Irene, Upper and Lower Kitandara) that are located near or above 4,000 m (3,890–4,487 m) in the Alpine and Nival zones, and receive at least some direct input of glacial meltwater. These lakes are mildly acidic to pH-neutral (surface pH: 5.80–7.82) and have above-average transparency (Secchi depth: 120–280 cm); their DOC content is noticeable but modest (3.0–5.5 mg/l), and epilimnetic NO3 concentration is always well above the detection limit (0.2–2.9 μg/l). These lakes have very low to mid-range TP (3.1–12.4 μg/l), TN (155–393 μg/l) and Chl-a (0.3–10.9 μg/l), and their range of TN/TP values (22.9–81.4 by mass) suggests that primary production is phosphorus-limited.

The second group of 11 lakes are mostly located below 4,000 m (2,990–4,054 m), more remote from the glaciers and surrounded by Ericaceous vegetation or, when in the Alpine zone, fringed by a Carex-Sphagnum bog. These lakes are mildly to strongly acidic (surface pH: 4.30–6.69), and less transparent (Secchi depth: 60–132 cm) usually because they are stained by high concentrations of dissolved organic carbon (DOC: 6.8–13.6 mg/l). The exception is Lake Mahoma at the Forest/Bamboo transition, which is acidic (pH: 5.20) and modestly transparent (Secchi depth: 82 cm) but less burdened with humic acids (DOC: 5.5 mg/l) and has a greenish colour. In almost all lakes of this second group including Mahoma, both PO4-P and NO3 concentrations are below the detection limit of 0.1 μg/l. Compared to the first group, these lakes have higher epilimnetic concentrations of TP (10.2–68.4 μg/l) and TN (213–696 μg/l), but their range of Chl-a (1.0–10.6 μg/l) is similar.

All sampled Rwenzori lakes have high to very high concentrations of suspended organic matter (POC range: 0.72–4.13 mg/l; PON range: 0.08–0.34 mg/l; see Electronic supplementary material), with no marked difference between the two groups of lakes defined above. The carbon-to-nitrogen ratio (C:N) of particulate organic matter (POC/PON), with ranges from 8.3 to 12.1 with a mean value of 10.5.

Discussion

Thermal and stratification regimes

The temperature of mountain lakes depends on elevation (through its effect on air temperature), on the proximity of a glacier (specifically, whether or not the lake is directly fed by glacial meltwater), on the number of hours of direct insolation, and on their exposure to wind. In stratifying lakes, bottom temperatures integrate at least the daily variation in surface-water temperature, and are therefore more representative of mean local conditions. The observed strong relationship between bottom temperature and elevation in our lake data set (r = − 0.84, P < 0.001) mirrors the highly significant correlation between mean annual air temperature and elevation in the Rwenzori (r = − 0.98, P < 0.001; Eggermont et al., unpublished temperature logger data), indicating that surface air temperature is a key factor in determining the temperature regime of Rwenzori lakes. The scatter that does occur in the relationship between lake temperature and elevation is nevertheless significant, and points to substantial influences of lake hydrology and exposure on local thermal regimes. For example, Upper Kitandara Lake is colder than other lakes at similar elevation, likely because it is fed directly by glacial meltwater from Mts. Baker and Stanley, and furthermore located in a narrow valley tunnelling cold wind from the glaciated peaks. In other lakes, deep wind-driven mixing during periods of incipient surface heating can raise their bottom temperature above the value predicted from its relationship with elevation. Such lakes (e.g., Lower Kachope; Table 1) also tend to feature modest surface warming and a weakly pronounced thermocline. Latent heat losses from the Rwenzori mountain lakes through evaporation must be generally minimal, due to the prevailing cool and humid conditions.

Hutchinson and Löffler (1956) and Löffler (1957, 1960, 1964) described lakes of tropical high-mountain regions in Africa and South America as frequently (even daily) stratifying and mixing at temperatures below 10°C, and therefore designated all of them as ‘cold polymictic’. Ruttner (1963) adopted their reasoning when assigning this term to all lakes that are not ice-covered and circulate daily at temperatures at or slightly above 4°C. Lewis (1983) noted the arbitrary nature of the 10°C boundary, and presented an amended classification of mixing regimes which resolved this and other problems with the Hutchinson-Löffler scheme (Wetzel, 2001). In the Lewis (1983) classification, cold-monomictic lakes stratify because surface temperature drops well below 4°C (often resulting in seasonal ice cover) and no longer mix with the heavier, but warmer, bottom water. Depending on wind stress during the ice-free season, such lakes have bottom temperatures at or below 4°C for much of the year. When such lakes also stratify for short or longer periods during the ice-free season by surface heating above 4°C, they are designated as cold polymictic. The Lewis (1983) classification is still somewhat unsatisfactory for tropical high-mountain lakes, which can briefly freeze over at night but during daytime are usually ice-free year-round (Löffler 1964). In terms of the ambient temperature regime for aquatic flora and fauna, the most important differentiation here is between mono- or polymictic lakes that stratify due to surface heating (and thus have hypolimnia of 4°C or higher) and mono- or polymictic lakes that stratify due to surface cooling (and thus have hypolimnia of 4°C or cooler). In this respect, all 17 Rwenzori lakes we surveyed so far are warm stratifying. Most lakes in the Alpine Zone (Lac du Speke, Upper Kitandara, Kopello, Bigata, Africa) have bottom temperatures slightly above 4°C (range 4.8–5.5°C). Lake Mahoma at 2,990 m has a bottom temperature of 12.5°C, a clear reflection of the higher mean annual air temperature at this lower elevation. The intermediate bottom temperatures of lakes in the Ericaceous zone (range 6.1–7.5°C; n = 7) must at least partly also be due to the influence of ambient air temperature, but can also result because deep wind-driven mixing continues into periods of incipient surface heating. Cold-stratifying mountain lakes, although not represented in our data set, do occur in tropical Africa: Lewis Tarn and Curley Pond on Mt. Kenya, and Lac Gris and Lac Blanc on the drier DR Congo side of the Rwenzori all have bottom temperatures well below 4°C (data from Löffler (1964) and de Heinzelin & Mollaret (1956).

Based on our current data set, lakes on the Uganda side of the Rwenzori broadly classify into three types of mixing regime. In Type 1 (Lake Katunda, Fig. 6a), uniform temperature, oxygen, pH and conductivity throughout the water column except the very surface suggests that it undergoes a continuous cycle of weak day-time stratification and night-time mixing. Given the absence of seasonal ice cover, this lake classifies as continuous warm polymictic. Type 2 groups lakes characterized by a more or less well-developed primary thermocline at the depth of wind-driven mixing (here 3–5 m depth), and modest to strong oxygen depletion in the hypolimnion, however without it becoming completely anoxic. This suggests that these lakes stratify for at least several weeks or months at a time, alternating with periods of mixing when fresh oxygen is supplied to the hypolimnion. These lakes classify as discontinuous warm polymictic. Evidently, a more productive Type 2 lake that mixes frequently may better prevent hypolimnetic oxygen depletion than a less productive lake mixing rarely, consequently the magnitude of oxygen depletion observed at any given moment (Fig. 6b, d) is a poor guide to the frequency of mixing. Some Type 2 lakes with either modest or strong oxygen gradients may even circulate only once each year, and hence classify as warm monomictic.

Accurate determination of mixing frequency in individual Rwenzori lakes will require year-round water-column monitoring. At this time we differentiate between Type 2a and Type 2b lakes because the relative magnitude of oxygen depletion has a significant influence on benthic habitat quality, and thus potentially on benthic invertebrate species distribution. Finally, Type 3 lakes (Bigata, Mahoma; Fig. 6e, f) are characterized by a strongly clinograde oxygen curve and completely anoxic bottom waters with significantly higher conductivity than surface waters. This suggests that stratification persists throughout the year, due to a chemical density difference between bottom and surface waters. These two lakes are meromictic, i.e. they lack at least one complete circulation of the water column per year. The cause of their chemical density gradient is probably biological, i.e. a gradual accumulation over many years of chemical substances resulting from the decomposition of organic matter (Wetzel, 1983). In the case of Mahoma, at least, the onset of meromixis sometime in the past may have been promoted by wind shelter in its steep-sided catchment.

Factors influencing transparency

The Secchi-depth transparency of the Rwenzori lakes (0.60–2.80 m; mean 1.35 m) is on average distinctly lower than that of 14 alpine lakes on Mt. Kenya surveyed by Löffler (1968a), which had a mean Secchi depth of 3.80 m (range 1.0–10.0 m). Tight coupling between DOC and Fe, and a significant inverse relationship between Secchi depth with both DOC and Fe, in the Rwenzori lakes indicates that high dissolved Fe concentrations are stabilized by organic complexing substances (humic acids), and that their presence is the principal regional control on water-column transparency. As evident from the PCA, these three variables were all strongly tied to the type of catchment vegetation. The least transparent lakes (SECCHI <1.00 m: Nsuranja, Africa, Batoda and Kanganyika) are all surrounded by lush Ericaceous vegetation and/or sizable bogs. The most transparent lakes (SECCHI >1.80 m: Upper and Lower Kitandara, Bujuku) are located in the Alpine zone near glaciers, lack Carex bogs, and have low DOC (<4 mg/l) and Fe (<0.1 mg/l) contents.

This differentiation between lakes with different types of catchment vegetation also showed from the RDA. The Nival and Alpine location of lakes without bogs each contributes significantly and independently to explained variation in the lake data set, as do the combined group of Alpine and Ericaceous locations with bogs. RDA groupings of all Ericaceous zone locations with and without bogs, or all Alpine zone locations with and without bogs, do not independently contribute to explained variation in the data set, pointing to the decisive influence of the Carex-Sphagnum bogs on the aquatic environment.

DOC is a key controlling factor in the underwater penetration of solar ultraviolet radiation (UV) (reviewed by Vincent & Pienitz, 1996; see also Pienitz and Vincent, 2000). DOC more strongly absorbs the biologically damaging UV-A radiation (Scully & Lean, 1994), and is thus a natural sunscreen for aquatic biota. Very likely, lack of this UV protection in the more transparent Rwenzori lakes explains the presence of Chydoridae with strongly pigmented carapaces in some of them (Eggermont et al., in press; see also Hansson, 2004; Tollrian & Heibl, 2004).

In glaciated mountain ranges elsewhere, low lake-water transparency is often caused by high loads of suspended inorganic solids in the glacial meltwater passing through them. In comparison, Rwenzori glaciers carry very little sediment on their surface, and the streams flowing out from beneath the glaciers are clear, not turbid as typical meltwater streams (Busulwa & Bailey, 2004). Glacially-derived sediment input certainly does affect some of the Rwenzori lakes (e.g., Bujuku with its well-developed delta and silty shorelines), but overall a lake’s proximity to the glaciers does not seem to exert a major influence on water-column transparency.

Factors influencing water chemistry and pH

Rwenzori lake-waters are primarily derived from surface runoff, and only secondarily from direct rainfall and glacier melting. Hence, bedrock composition can be expected to exert dominant control on lake chemistry. In the adjacent lowlands of western Uganda, most lakes occur on volcanic bedrock (Melack, 1978), and receive high inputs of Na, Mg and K relative to Ca (Fig. 7a). High weathering rates in the sub-humid tropical climate there produce large amounts of HCO3 (Fig. 7b), and evaporative concentration of dissolved solids often leads to biologically induced precipitation of carbonate, causing further loss of Ca. Bedrock in the Rwenzori is more variable. The highest peaks are composed of volcanic amphibolite (Osmaston & Pasteur, 1972). Below it occur softer schists, and at lower elevations the range is primarily composed of weathered Precambrian gneiss. The somewhat higher conductivity of lakes in the Bujuku and Butawu river systems (18–52 μS/cm; n = 7) likely reflect the weathering of amphibolite and schists exposed on high peaks nearby. The alpine lakes in this group (Bujuku, Upper and Lower Kitandara) also have elevated SO4 concentrations. The lower conductivity of lakes in the other river systems (5–21 μS/cm; n = 8) reflects slow weathering of the underlying gneiss. As expected, the lowest conductivity values (5–8 μS/cm) occur in lakes (Irene, Ruhandika) and pools in the Nival zone that are surrounded by bare rock. Thus, factors contributing to the extremely dilute character of Rwenzori lake waters, besides slow-weathering bedrock and low temperatures, are the steep mountain slopes which allow percolating water to pass quickly, and the near-continuous rainy weather, which dilutes the scarce ions released by weathering. Given the extremely low conductivity of most Rwenzori lakes, salts carried by rain (Na, Cl and SO4) might have a noticeable influence on their hydrochemistry. But in fact, the average concentrations of conservative ions in the lakes (see Electronic supplementary material) are even lower than values reported for lowland rainfall (Visser, 1961). This is likely due to the great distance to the Indian Ocean, and efficient rain-out of sea-spray salts before reaching Rwenzori’s slopes.

The cation concentration sequence in Rwenzori lakes and pools surveyed in this study (Ca > K > Na > Mg) is strongly biased toward products of feldspar weathering, as in Congolese Rwenzori lakes (de Heinzelin & Mollaret, 1956) but deviating from that in the lowland lakes of western Uganda where Mg, Na and/or K concentrations often surpass Ca. Also the anion pattern (the SO4 > HCO3 > Cl and SO4 > Cl > HCO3 sequences occur in about equal proportion) deviate from the HCO3 > Cl > SO4 sequence found in most lowland Uganda lakes (Melack, 1978; Kilham, 1971; Kizito et al., 1993; Verschuren et al., unpubl. data).

The acidity of many Rwenzori surface waters is primarily due to the presence of humic substances leached from organic top soil and the bogs (viz. the significant inverse relationship between DOC and pH). The circum-neutral pH of lakes in the Bujuku and Butawu drainages (6.55–7.81; n = 7), including DOC-stained lakes surrounded by lush Ericaceous vegetation and/or Carex-Sphagnum bogs (Bujuku, Middle Kachope), must primarily derive from the buffering capacity of their catchment’s rock geochemistry rather than CO2 uptake by phytoplankton.

Nutrients and primary production

Surface-water TP concentrations in Rwenzori lakes (3.1–68.4 μg/l) are high on average compared to mountain lakes in Europe (2–7 μg/l; Psenner, 1989; Marchetto et al., 1994; Camarero et al., 1995; Mosello et al. 1995; Skjelkvåle & Wright, 1998; Kopáček et al., 2000), North America (2.5–15 μg/l; Schindler, 2000) and Asia (average 7.3 μg/l, Lacoul & Freedman, 2005). Lakes in the Nyamughasani and Nyamwamba valleys, in particular, contained very high TP levels (>20 μg/l). However, since PO4-P in all surveyed lakes was below detection limit, and TP is positively correlated with DOC, it follows that most of the measured TP must be biologically unavailable phosphorus bound to allochthonous organic matter. Lakes with a POC:Chl-a ratio greater than 100:1 likely derive most of their POC input from terrestrial or detrital sources (Eppley et al., 1977). POC:Chl-a ratios in the Rwenzori lakes (85–2,471; mean 625:1) are decidedly extreme, again indicative that most organic matter suspended in the water column is produced from dead terrestrial vegetation rather than the result of autochthonous production. This conclusion is further supported by the carbon-to-nitrogen ratio (C:N) of particulate organic matter (POC/PON), with ranges from 8.3 to 12.1. Organic matter produced by algae tends to have a C:N ratio of 6–9, while vascular land plants produce organic matter with C:N ratios of 20 and greater (Meyers & Teranes, 2001). Wetland soils, and Sphagnum-dominated wetlands in particular, tend to have C:N values of 9–17, i.e. lower than typical terrestrial organic matter and only slightly above those of lacustrine organic matter (Meyers & Teranes, 2001). The C:N values recorded here can be interpreted to represent organic matter derived primarily from the lakes’ catchments, particularly wetlands soils. This is confirmed by both PCA (Fig. 2) and RDA ordinations, which showed total nutrient concentrations to be strongly linked to the type of surrounding land cover, with the lowest values recorded in lakes surrounded by bare rock or alpine vegetation without bogs, and the highest values recorded in lakes surrounded by Ericaceous vegetation and/or bogs.

Lake productivity classifications on the basis of epilimnetic TP, TN and Chl-a concentration (Forsberg & Ryding, 1980; Downing & McCauley, 1992) identify the group of six high-alpine, transparent Rwenzori lakes as ultra-oligotrophic (Irene) to mesotrophic (Ruhandika) and phosphorus-limited (TN/TP by mass: 23:1–81:1; PO4-P below detection limit; NO3: 0.2–1.4 μg/l). In the humic, DOC-stained lakes at lower elevation, autochthonous primary production must be limited both by phosphorus and light penetration. As both PO4-P and NO3 in these lakes are below detection limit, nitrogen limitation may also affect algal community structure. However the very low TN:TP ratios in some lakes (9:1 to 12:1 in Kopello, Africa and Kanganyika) more likely results from the pre-dominant contribution of dead, allochthonous organic matter to total nutrient concentrations (McNeely et al., 1979).

Summary and perspectives

In the context of ecological and biodiversity research, aquatic habitat in mountain lakes on the Ugandan side of the Rwenzori range is structured along two major environmental gradients. With regard to the limnetic habitat, lakes differentiate between (1) slightly acidic to pH-neutral, phosphorus-limited clear-water lakes near or above 4,000 m elevation, with at least some direct input of glacial meltwater and surrounded by alpine vegetation or rocky catchments; and (2) more strongly acidic, light- and phosphorus-limited humic lakes mostly below 4,000 m elevation, remote from the glaciers and surrounded by Ericaceous vegetation and/or bogs. With regard to benthic habitat, the Rwenzori lakes surveyed to date have mixing regimes ranging from polymictic to meromictic, resulting in hypolimnetic (profundal) oxygen regimes ranging from near-saturated over mildly depleted to completely anoxic, despite generally low primary productivity. Excluding the mid-elevation Lake Mahoma, the measured range in surface-water temperature among Rwenzori lakes (5.5–9.1°C) is relatively modest, overlapping substantially with the range of bottom-water temperatures (4.8–7.5°C). Hence, we do not expect the benthic fauna of well-oxygenated profundal habitat to be markedly different from that on similar substrate in shallow-water habitat. We do expect species turnover in profundal benthic communities of Rwenzori lakes along the oxygen gradient controlled by mixing frequency.

Alpine lakes in remote and undisturbed regions are known early-warning systems for more widespread environmental change (e.g. Schmidt & Psenner, 1992; Sommaruga-Wögrath et al., 1997; Skjelkvale & Wright, 1998; Battarbee et al., 2002; Rogora et al., 2003). As shown from this study, the physical limnology, hydrochemistry and nutrient budget of Rwenzori lakes varies closely with climate-controlled characteristics of their local abiotic environment, in particular air temperature, proximity to glaciers and type of vegetation cover. It follows that Rwenzori lake ecosystems may indeed prove to be very sensitive to anthropogenic climate change and glacier retreat.

In temperate regions, air temperature is recognized to influence the timing of ice break-up and growing season in lakes (Livingstone, 1997; Palecki & Barry, 1986). Climatic warming in the Rwenzori may enhance the thermal stratification of lakes and thus reduce their frequency of complete water-column mixing and aeration. Combined with high inputs of terrestrial organic matter, this may result in more widespread occurrence of low-oxygen conditions in the hypolimnion, profoundly changing the biogeochemistry and aquatic community structure of these lake systems. Rising regional air temperature will also likely result in a gradual upward shift of vegetation belts (Beniston, 2005; Walther et al., 2005), which will in turn affect the lakes’ nutrient budget and DOC content, water-column transparency and UV penetration. The low buffering capacity of most Rwenzori lakes also makes them vulnerable to the acidification associated with incomplete organic decomposition (see Psenner & Schmidt, 1992; Schmidt & Psenner, 1992; Wright & Schindler, 1995; Koinig et al., 1996). Glacier retreat will cause a larger proportion of the catchment to be exposed to physical and chemical weathering (Wright & Schindler, 1995; Skjelkvale & Wright, 1998) changing nutrient and ion budgets accordingly. In comparison, the thermal effects of changes in glacial meltwater input may be modest, except in lakes now in close proximity of the remnant glaciers.

The limnological data resulting from this study may serve as an important quantitative baseline for long-term environmental monitoring of the Rwenzori lakes. The exact amplitude and direction of ecosystem change due to climatic warming, however, depend on site-specific processes and are not readily predictable (Psenner & Schmidt, 1992; Koinig et al., 1998). Use of these limnological data to create paleoenvironmental calibration data sets will, when applied to sedimentary records, help to improve insight in the long-term natural variability of these ecosystems and in their actual sensitivity to climate-driven environmental change.

Acknowledgements

The fieldwork was conducted under Uganda NCST research clearance NS21 and Uganda Wildlife Authority permit UWA/TBDP/RES/50, with logistic support from Rwenzori Mountaineering Services. We greatly thank Halewijn Missiaen and Bob Rumes for field assistance, Angelica Alcantara for lab assistance, Renaat Dasseville for pigment analyses (at Ghent University, Belgium) and Lei Chou for TP and TN analyses (at the Univerisité Libre de Bruxelles, Belgium). Elie Verleyen is acknowledged for constructive comments on the statistical analyses, and Steven Declerk and two anomymous referees for supportive critique which helped to improve the manuscript. This research was sponsored by the Fund for Scientific Research of Flanders, the Leopold III-fund Belgium (Belgium), the Stichting Ter Bevordering van het Wetenschappelijk Onderzoek in Afrika (Belgium), and the U.S. National Geographic Society (grant 7999-06). H.E. is a postdoctoral fellow with the Fund for Scientific Research of Flanders (FWO-Vlaanderen).

Supplementary material

10750_2007_Article_741_ESM.pdf (444 kb)
ESM (PDF 445 kb)

Copyright information

© Springer Science+Business Media B.V. 2007

Authors and Affiliations

  • Hilde Eggermont
    • 1
  • James M. Russell
    • 2
  • Georg Schettler
    • 3
  • Kay Van Damme
    • 1
  • Ilse Bessems
    • 1
  • Dirk Verschuren
    • 1
  1. 1.Limnology Unit, Department of BiologyGhent UniversityGentBelgium
  2. 2.Department of Geological SciencesBrown UniversityProvidenceUSA
  3. 3.Section Climate Dynamics and SedimentsGeoForschungsZentrum PotsdamPotsdamGermany

Personalised recommendations