Image assessment of water hyacinth infestation in Lake Victoria, East Africa
Joel Masselink, GISDE Graduate Student, Clark University, firstname.lastname@example.org, December 2007
Project Purpose In the late 1990s, bays across Lake Victoria, the second largest lake in the world, were paralyzed by massive blooms of water
hyacinth, an invasive water-borne flower. This infestation was of massive scale, but was quickly alleviated by both biological and
human factors. So what has happened since? The study period of 2000 to 2003 was chosen because of the scarcity of research following the alleviation of the
Data Collected horrible late 1990s infestation.
Water hyacinth is an invasive species that first appeared in the lake in the late 1980s. It thrives in vulnerable ecosystems and cause a major nuisance. The
Landsat Scenes - Path 170, Row 60 aggressive growth pattern causes problems for transportation, fisheries, hydropower, and water intake systems. Hyacinth disturbs aquatic ecology by depleting
the oxygen content of the water and changing interaction between species. Thick mats of vegetation provide cover for smaller fish, and push larger fish further
TM: 1986; ETM+: 2000, 2001, 2002, 2003 from the shallows due to lack of oxygen. Infestations could structurally alter ecosystems if left unchecked. This study will assess post-outbreak persistence.
Technical This project uses available image data to assess distribution of water hyacinth in the Winam Gulf of Kenya from 2000-2003.
Complementing this study period were freely available, high-quality Landsat images. A multispectral scene was collected for 1986
Procedure and each year of the study period. Raw scenes were imported, assessed for quality, and windowed to matching extents.
Characterization of water-borne vegetation required separation of land and water, to avoid misclassification of vegetation. The lake extent was calculated by
comparing image bands 5 from 1986 and 2001 and band 7 from 1986. Water is characterized by very low reflectance values in bands 5 and 7, which record Winam Gulf, Kenya
reflectance in the mid-infrared range (1.55-1.74 µm and 2.08-2.35 µm respectively). The 1986 Landsat TM scene was used to calculate the lake’s surface because
it was taken before record of hyacinth in the lake. Band 5 from 2001 was carefully chosen for its absence of water-born vegetation. A Boolean mask was created to Lake Victoria
eliminate land from the assessment. This mask is valuable for image overlay calculations allowing for quantification and analysis of water-borne vegetation. The Winam Gulf is the main Kenyan portion of Lake Victoria. According to a
NDVI was calculated for each time to characterize vegetation using the NDVI formula of (NIR-R) / (NIR+R). Positive NDVI was assumed to represent live comprehensive lake-wide study for 1989-2001, this gulf was covered with over
vegetation. NDVI values were reclassified to Boolean representing vegetation and none. These images were used for comparison calculations. Clouds induced 17,000 hectares of hyacinth in November 1998 (Albright et al.), the very worst of
error by causing uncertain positive NDVI values in 2003, so parameters were changed to cause maximum elimination of unknown NDVI and minimal the lake wide infestation.
misclassifications of water-borne vegetation.
This study focuses on the cities of Kisumu and Kusa. Kisumu is the third largest
The Boolean NDVI images were aggregated into groups of contiguous vegetation and assigned area in hectares. Groups with area greater than 10 hectares
city in Kenya, with a population of over 300,000. Most commerce is directly
were considered significant, because mats that size would be a nuisance. This threshold of 10 hectares eliminates unwanted noise and error associated with
related to the advantageous lake front location. Kusa is located near the mouth of
misclassification of the lake and clouds. The three images depicting groups were compared in chronological order, producing three comparison images for:
Image Processing Model 2000:2001, 2001:2002, and 2002:2003. These were compared in a three image crosstab to show persistence throughout the entire study period.
the river Nyando. This confluence forms a sizeable wetland habitat characterized
by papyrus stands. Papyrus is valuable for making mats, and the wetland habitat
is favorable for rice cultivation, so the stands are disappearing.
2000/03/06 Population and industry around the lake have grown rapidly in the last few
2001/02/05 decades. Today, Lake Victoria supports a population of around 30 million people
2002/03/28 whose livelihood is largely dependent on local resources. The vulnerability of this
2003/05/18 Cluster ecosystem is cause for serious concern.
Band 5 Rank Persistence
Crosstab & Reclass
2001/02/05 7 All 4 images
Window Overlay (*) & Rank & Compare
Band 5 6 3 in a row
1986/03/08 5 3 of 4
Band 7 4 2 in a row
3 2 of 4
MASK1 = Crosstab 4 = Vegetation both
2 1 of 4
Windowed Bands 3 & 4 Land Mask Comparing 3 = Veg later time
1 0 of 4
for all study years Boolean NDVI Consecutive 2 = Veg earlier time
Overlay Overlay NDVI Grouped NDVI 0 Land Mask
Masked images of Reclass Group & Area Only hyacinth Years 1 = No vegetation
Band * MASK1 Vegetation = 1
all study years NDVI (4-3) / (4+3) NDVI > 0
None = 0 >10 ha mats > 10 ha
Results Map 1. Persistence 2000-2003 Figure 1. Area affected by Vegetation Interpretation of Results What can prevent future infestations?
Water level, climatic conditions, siltation, pollution and biodiversity
Hyacinth in Study Area are factors in the prevention equation. Relief from the 1998 outbreak
Kisumu and Kusa experienced most of the Any explanation for spatial trends requires analysis of was largely credited to predatory weevils and El Nino induced winds,
observed hyacinth persistence during the study 800 climate and environmental policy factors. but sustained hyacinth control requires adaptive management of
period. As shown in Map 1, significant concentrations 700 Highest quantities were measured in 2002, while the several factors.
were found in Kusa for every time and in Kisumu all 600 most mats appeared in 2000 and 2003. Defining a cause
but one of the times assessed. Kusa’s bay had an 500 for these macro-scale trends is beyond the scope of this
Lake Level compared to 10 yr Mean
average total of 285 hectares each time. Seven project.
hectares persisted through all four times. Vegetation
clogged Kisumu’s bay in 2000, vanished in 2001 and 300 Measurable factors include rainfall and water level, 0.5
large mats persisted for 2002 and 2003. 200 which is also a function of power demands. These have
Table 1 shows the quantities found in key locations 100 caused hydro controllers to release the reservoir for power Time 1
at each time. The quantities of vegetation 0 generation, which accounts for changes in the lake level. Time 4
2 3 0.2
persistence associated with Map 1 are shown in 1 tim e 2 times
3 of 4 times
Land use should also be considered, as hyacinth has 0.1
Figure 1. been linked to low-water quality and siltation.
Area (Ha) 718.84 155.55 96.98 43.21 109.82 6.82
Two-thirds or 1130 of 1710 total observed hectares Persistence
of hyacinth accounted for original incidence, while the
remaining third or 580 hectares represented hyacinth
Each study time had an average of 427.5 hectares
Table 2. Mat size per study time -0.4
of hyacinth. Area of Mat size is located in Table 2. Mats (area ha) 2000 2001 2002 2003 Total -0.5
Study Pe riod
Quantifying hyacinth concentrations found at these 177 570 153 261
locations may provide an idea as to the scale of 157 72 108 50
overgrowth and persistence that affected this Sources: USDA Foreign Agriculture Service PECAD, NASA Earth Observatory
31 21 21
population. Table 1. Location of Vegetation 15 18 Acknowledgments: IDCE 371 classmates, Dr. Ogneva-Himmelberger, Benoit Parmentier, Dr. Les Kaufman, Clark Labs
14 17 Citations: Albright, T. P., T. G. Moorhouse, and T.J. McNabb. 2004. The rise and fall of water hyacinth in Lake Victoria and the
Location 2000 2001 2002 2003 Avg
How big is a hectare? Kisumu 192 0 108 50 70
13 12 Kagera River Basin, 1989-2001. Journal of Aquatic Plant Management 42:73-84
2.47 acres, 100 by 100 meters, or the size of Clark’s main quad. COUNT 6 2 3 6 17 Data providers: Clark Labs, Global Land Cover Facility –University of Maryland Institute for Advanced Computer Studies,
Kusa 157 570 153 261 285 TOTAL AREA 407 642 282 379 1710 Intergovernmental Authority on Development Livestock Policy Initiative:
How big was the largest mat? Other 58 72 21 68 55 AVG SIZE (ha) 67.83 321.00 94.00 63.17 137 GLCF – UMIACS: http://glcfapp.umiacs.umd.edu:8080/esdi/index.jsp
570 hectares, which is the size of an 18 hole golf course. Total 407 642 282 379 246 IGAD LPI Online Data Portal: http://ergodd.zoo.ox.ac.uk/igadweb/tiki-index.php
Map Library: http://www.maplibrary.org/stacks/Africa/index.php