An Analysis of Hydrological Changes in Northwestern Lake Victoria, Tanzania

Report by Thomas A. Woolman

ES775 - Advanced Image Processing, Emporia State University

Fall Semester, 2013

Figure 1: Map of central Africa, with Bukoba, Tanzania on the western shore of Lake Victoria marked as Point A


Introduction Vegetation Analysis of Deforestation and Agricultural Development | NDVI - Normalized Difference Vegetation Index |
Band Ratio Overlay Composite Imagery Analysis | Cluster Analysis - Unsupervised Investigation
Analysis and Conclusions | References


Lake Victoria (Victoria Nyanza in Bantu Swahilli) is one of the African Great Lakes, named after Queen Victoria in 1858. It has a surface area of 68,800 square kilometers making it the largest African lake by surface area, the second largest freshwater lake in the world by surface area, as well as the largest tropical freshwater lake in the world. See Figure 1 above for a map of Central Africa showing the relative size and significance of Lake Victoria. Bukoba, Tanzania along the northwestern shore of the lake is marked as Point A.

Lake Victoria receives its water primarily from direct precipitation and thousands of small streams. The largest stream flowing into this lake is the Kagera River flowing from northwestern Tanzania near the Rwandan/Uganda border, the mouth of this river lies on the lake's western shore. Two rivers leave the lake, the White Nile (known as the "Victoria Nile" as it leaves the lake), flows out at Jinja, Uganda on the lake's north shore, and the Katonga River flows out at Lukaya on the western shore connecting the lake to Lake George (Bossche, Bernacsek 1990). See Figure 2 below for a more detailed map showing the interdependence between multiple East African nations who all largely rely on Lake Victoria for food sustenance.

Figure 2: Detailed map of the northwestern Tanzania with Rwanda, Uganda, Burundi and Kenya shown.
Bukoba, Tanzania is marked as Point A.

Lake Victoria has a maximum depth of 84 m and an mean depth of 40 m, with a shoreline of 4,828 km which includes several hundred large and small islands accounting for 3.7% of the shoreline. Its surface area covers three countries, Kenya at 6% of the lake, Uganda at 45% of the lake and Tanzania at 49% of the lake (Prado, et al., 1991) It contains approximately 2,750 km2 of water.

Because Lake Victoria is extremely shallow relative to its overall large surface area, it is extremely vulnerable to the effects of stream flow and drainage modification, hydrology changes brought about by human habitation impact, and percipitation patterns. Agricultural practices including irrigation and chemical fertilizer pollution have been a major cause of problems for the lake. Additionally, deforestation for fuel and subsequent erosion especially after the Rwandan Genocide which began in 1994 that prompted massive migrations out of Rwanda into northwestern Tanzania have caused dramatic changes to the hydrology of the lake. Hydroelectric projects in Uganda after the construction of the Nalubaale Dam in 1999 and the Kiira Dam in 2006 for the generation of electric power have been the largest cause of drainage from the lake, estimated at over 2 meters of depth lost since 2006 (Stuteville, 2008).

This report will analyse some of the effects of the combination of those factors to the overall size, appearance and health of the lake, using data from NASA Landsat remote imaging satellite systems. A wide range of techniques will be utilized including natural and false color composite datasets, NDVI vegetation index analysis and unsupervised cluster analysis to attempt to quantify land changes that may have impacted the hydrological cycle of the lake. Landsat image datasets will be comprised of an initial baseline dataset from Landsat 5 dated 31 July 1984 and a final dataset from Landsat 8 dated 29 June 2013. Both datasets were obtained in the June-July timeframe of their respective years, after the completion of the "long rains" season, which averages from 25 March to 21 May each year. See Figure 4 below for a comparison of the bands and their resolution with utilizations for both of these two satellites.

Of interest in these map datasets is the town of Bukoba, a town in northwest Tanzania on the western shore of Lake Victoria (see Figure 3). It is the capital of the Kagera region, and the administrative seat for Bukoba Urban District. The urban population is estimated at 100,000. Situated on the shore of Lake Victoria, Bukoba lies only 1 degree south of the Equator and is Tanzania's second largest port on the lake. The regional capital and Tazanian's Kagera district's biggest town is the gateway to the northwestern lake region. The climate is sunny and mild most of the year. The town is flat and compact, forming a bowl as it is surrounded by hills (Wikipedia, 2013).

Figure 3: Natural Color Composite Image from Landsat 5, 1984 (TM bands 123)
showing Bukoba, Tanzania and surrounding countryside and shoreline, for baseline comparison.

Landsat 8 OLI and Landsat 5 TM scanner comparison
Landsat 8
OLI band (m)
Landsat 5
TM band (m)
Color Primary use
1. 0.4330.453
Violet-blue Shallow water, aerosols
2. 0.4500.515
1. 0.45 - 0.52
Blue-green Shallow water, land cover
3. 0.5250.600
2. 0.52 - 0.60
Green Vegetation, land cover
4. 0.6300.680
3. 0.63 - 0.69
Red Minerals, soils, land cover
5. 0.8450.885
4. 0.76 - 0.90
Near-infrared Vegetation, NDVI
6. 1.5601.660
5. 1.55 1.75
Mid-infrared Soil moisture, minerals
6. 10.4 - 12.5
Thermal-infrared Surface temperature
7. 2.1002.300
7. 2.08 - 2.35
Mid-infrared Soil moisture, minerals
8. 0.5000.680
Panchromatic Image sharpening
9. 1.3601.390
Mid-infrared Cirrus clouds

Figure 4: Table showing comparative bands between Landsat 8 and Landsat 5 and their relative utilizations

Vegetation Analysis of Deforestation and Agricultural Development on Lake Drainage

Deforestation is a growing problem facing northwestern Tanzania and this accelerated after the 1994 Rwandan Genocide which caused a massive influx of refugees into Bukoba and the surrounding Kagera region of Tanzania. "Deforestation is caused by overgrazing and shifting cultivation. Overgrazing mainly occurs in Biharamulo and Karagwe districts as well as in some parts of Muleba. The situation is worsened by intensive clearing of forests for cotton and paddy (rice) farming. Woodfuel demand is also another cause of deforestation." (Environmental Information Centre of the National Environment Management Council, 1997).

"..the rate of clearing forests for agriculture is alarming. Public lands which were formerly stocked with big diameter trees are now left with bushes and shrubs. In recent years, when the 1994 Rwanda genocide occurred, hundreds of thousands of refugees crossed the border and entered the Region to seek safety." (Environmental Information Centre of the National Environment Management Council, 1997). This has had a demonstrative impact on the state of natural forestation and the increasing use of irrigated cropland agriculture, which significantly lowered the amount of water and water quality draining into Lake Victoria. In this case, Landsat 5 data incorporating TM bands 2, 4 and 5 (Visible Green, Visible Red and Near-Infrared) was utilized alongside Landsat 8 data incorporating near-comparable OLI bands 4, 5 and 6 (Visible Blue-Green, Visible Red, Near-Infrared) to conduct a land cover change analysis.

Figure 5: False Color Composite Image from Landsat 5, 1984 (TM bands 245)
and Landsat 8, 2013 (OLI bands 456) showing significant deforestation and agricultural expansion

The use of Visibile Green, Visible Red and Near-Infrared datasets can be useful for assessing qualitative data about the type of vegetation ocurring within an area. In this case, TM bands 2, 3 and 4 were utilized (Visible Green, Visible Red, and Near-Infrared) alongside near-comparable OLI bands 2, 4 and 5 from Landsat 8. Light red indicates inactive/fallow vegetation while dark red indicates active agricultural planted vegetation. The results are significant and lend strong evidence to support the assertion that massively increased planting and irrigation utilization has happened over the 19 year time span, which would reduce drainage flow into Lake Victoria. The results shown in Figure 6 below are in general agreement with the image pattern shown in Figure 5 above.

Figure 6: False Color Composite Image from Landsat 5, 1984 (TM bands 234)
and Landsat 8, 2013 (OLI bands 245) showing significant agricultural expansion

Normalized Difference Vegetation Index

NDVI (Normalized Difference Vegetation Index) graphical indicator techniques were also used for both the 1984 and 2013 datasets, and the resulting images below (see Figure 7) indicate the massive level of deforestation and replacement by paddy crops and cotton fields. The NDVI values rank from -1 to +1. Negative values (if present) would represent inactive vegetation, while positive values represent active vegetation, generally in increasing scope of biomass. The analytical image for 2013 indicates far less vegetation biomass, indicative of rice paddies, sparse cotton farms and bananna plantations as compared to the previously existing natural forests. This has resulted in less runoff as compared to natural forest land and thus reduced stream flow to Lake Victoria.

Figure 7: NDVI Vegetation Index images from Landsat 5, 1984 (TM bands 34)
and Landsat 8, 2013 (OLI bands 45) showing massive deforestation

Band Ratio Overlay Composite Imagery Analysis

This set of images (see Figure 8) was created from the 1984 and 2013 datasets respectively. The Landsat 5 dataset from 1984 depicts a false color composite image with TM bands 1, 3 with an overlaid 7-4 as the third band. The 2013 datasat is from Landsat 8 and utilizes a false color composite of bands 2, 4 and an overlaid 7-5 for the third band. These set of bands and overlays are nearly-equivilant between the two satellites capabilities in terms of spectra and resolution. The analytical intent for these two composite images is to allow us to view the amount of surface water without being masked by surface vegetation. This would allow us to discern the effects of deforestation and the expansion of subsistence agriculture upon the hydrological drainage patterns of the lake shore region.

Free-standing water appears red in these images, and agricultural plantings appear as blue-green. Additionally, these images are both shown on a larger scale to show a fuller extent of environmental change that is apparently taking place. Utilizing the NDVI analysis that took place in Figure 6, we can discount any potential impact of seasonality variance (approximately 4 weeks of seasonal time difference between the 1984 and 2013 datasets) because the 1984 image in Figure 6, which takes place on 31 JUL 1984, has substantially more biomass indicative of natural forest than the 29 JUN 2013 dataset. There is a substantially larger amount of in-shore standing water, presumably related to subsistence agriculture (rice paddies, etc.) as well as a substantial increase in monoculture cash crops and a decrease in fallow fields and natural brushlands (pink/tan regions).

Proceeding further with our analysis, we can see that the increasing use of water-intensive cash crop and substitence agricultural practices have significantly increased since 1984, altering the hydrological drainage patterns which help to sustain Lake Victoria.

Figure 8: Band Overlay Composite Imagery. TM Bands 1 , 3 and Overlay 7-4 for the 1984 Dataset,
and OLI bands 2, 4 and Overlay 7-5 for the 2013 Dataset.

Cluster Analysis - Self-Organized Unsupervised Classification Analysis

Moving approximately 100 km south of Bukoba along the northwestern Tanzanian coast of Lake Victoria, I conducted an unsupervised classification analysis of the Kenmondo Bay and Bumbire Island region of Lake Victoria, to see if a similar pattern in the datasets emerge in a different area with a distinctly different analytical methodology. "ISOCLUST is an iterative self-organizing unsupervised classifier based on a concept similar to the well-known ISODATA routine of Ball and Hall (1965) and cluster routines such as the H-means and K-means procedures. Note that three modules are called by the ISOCLUST module: CLUSTER, MAKESIG and MAXLIKE. CLUSTER is used to derive the initial clusters, or seeds, from the set of input images. This result is then fed into an iteration of MAKESIG and then MAXLIKE." (IDRISI SELVA Help System, accessed 2013).

Using the ISOCLUST functionality within Idrisi Selva, I selected bands 2, 5 and 6 and 33 detected clusters from the Landsat 8 dataset, and near-equivalent bands 2, 4 and 5 from the Landsat 5 dataset which detected 32 clusters (see Figure 9 below). We can see a highly significant variance between the two datasets using this methodology, lending support to the assertion that deforestation and the implementation of susbsistence agricultural practices have greatly altered the hydrological drainage conditions into the lake. Of further interest is the gradiation of the lake waters itself detected in one of the bands in the 1984 dataset, indicating the possible presence of sedimentation in the lake drainage. This could be interpreted as an indication of a healthy lake environment as proper hillside drainage would be expected to carry minealization and organic compounds which would help sustain aquatic life. It's marked absence from the unsupervised classification algorithm's detection in the 2013 dataset is of interest and may be an indication of highly reduced drainage.

Figure 9: Self-Organized Unsupervised Classification Cluster Analysis, using TM Bands 2 , 4 and 5 for the 1984 Dataset
with 32 clusters, and OLI bands 2, 5 and 6 with 33 clusters for the 2013 Dataset.

Analysis and Conclusions

The evidence provided above indicates that significant human activities have occured over the nearly 30-year span between the two datasets. These activities have primarily been related to mass deforestation and with it the altering of hydrological patterns in the countryside, which have a direct influence on the water cycle that sustains Lake Victoria. In addition, the previously noted hydroelectric dams on the Uganda side of Lake Victoria have drained a significant amount of water since the Nalubaale Dam began operating n 1999, over and above their original design estimates. Increased migration into the lake region particularly after start of the Rwandan genocide of 1994 and an increasing population trend have significantly changed the landscape as demonstrated above. These factors are contributing to an overall decline in the health and sustainability of the lake as the dominant protein food resource for the people of central East Africa.


Environmental Information Centre of the National Environment Management Council, United Republic of Tanzania Government, 1997. "An Environmental Profile for Kagera Region, Tanzania" , Accessed 27 November 2013.

vanden Bossche J.-P., Bernacsek G. M., 1990. "Lake Victoria". Source book for inland fishery resources of Africa. Volume 1. CIFA Technical Paper No. 18/1. Food and Agricultural Organization. pp. 8387. ISBN 92-5-102983-0.

Prado J., Beare R.J., Mbuga Siwo J. & Oluka L. E., 1991. A catalogue of fishing methods and gear used in Lake Victoria. UNDP/FAO Regional Project for Inland Fisheries Development (IFIP), FAO RAF/87/099-TD/19/91 (En). Rome, Food and Agricultural Organisation.

Stuteville, 2008. "Lake Victoria beset by environmental problems", The San Francisco Chronicle, June 24, 2008. Accessed online 8 NOV 2013. , Accessed 26 November 2013.