Land Change and Desertification in the South Asembo Region and Lake Victoria
The South Asembo region along Lake Victoria in Kenya is one of the most underdeveloped, impoverished areas remaining in Africa. It is one of the hardest hit areas in terms of AIDS and Malaria, and faces constant issues with access to clean, safe drinking water. Pollution and runoff from livestock further pollute the limited water sources available, including the nearby Lake Victoria. Villagers in the region survive off of subsistence farming, often cutting down forested areas to access less degraded farming land. As a result of these practices, the South Asembo region has faced a good amount of land use and land cover change throughout the past, especially as population density increases around its shores.
Additionally, Lake Victoria itself has faced change through desertification and over-use of its resources. Desertification is the process by which a combination of climate and anthropogenic factors can lead to a drying up of vast areas of land or water bodies. Increased attention has been given to the declining water levels of Lake Victoria in recent years, as they have been shown to have decreased an average of at least six feet since 2001. Years of severe drought and changing climatic conditions have led to concern for the lakes of Africa, surrounding land, and rivers that feed off of these water bodies, such as the Nile River which begins its journey at Jinja, Uganda, the headwaters of the Nile. These waters, especially Lake Victoria, are ecological havens for a diverse array of plant and animal species that exist nowhere else in the world.
As these areas are faced with increased health and resource issues, it is important to look at the extent to which the land and water surfaces are changing over time, particularly in correlation with the changing population densities of the region. This project examines the land cover classification of the South Asembo region and shoreline of Lake Victoria in Kenya, particularly in correlation of the changing population density in the region. Two satellite images are examined; one from 1984 and one from 2008, to further examine the changes that occurred within the twenty-four year period. Remote sensing and the Imagine software allows for a large scale investigation and classification of the land in this remote area.
Satellite images were obtained from the United States Geological Survey (USGS) website, using the Glovis Global Visualization Viewer. I obtained approximate coordinates for the South Asembo area and used the search tool to find a list of available images. I selected two images with less than 10% cloud cover from 1984 and 2008 for this comparison. The images were analyzed using ERDAS Imagine Image Processing software.
To complete this analysis, I chose to do unsupervised classification of the two images, since I do not have enough familiarity with the region and land type to create the training areas necessary for a supervised classification model. ERDAS Imagine utilizes an ISODATA clustering technique to complete unsupervised classifications, meaning it uses the statistics of the image data to group the pixels into similar classes. For this project, I eventually decided on setting the model to create twenty classes with fifty maximum iterations to reach the convergence threshold of 0.950. This decision was made after repeatedly running the model with various values to determine which ones created the best result. Under “Initializing Options”, I chose the option to initialize means along the Principal Axis, rather than a forty five degree line.
The resulting classes were then identified and recoded based on aerial photography and general knowledge of the land area. I chose to separate the twenty defined classes into water, natural vegetation/forest, grasses/shrubs, bare soil, and agriculture/livestock pasture giving each a unique color scheme. A thematic recode was then performed to combine similar classes into five rather than twenty classes. At this point, it would have been ideal to perform a change analysis of the two images to quantify the differences between the two time periods, but due to a lack of resources this was not possible at this time. Additionally, there was a miscommunication with my former project partner about the images being rectified with true coordinates and it turns out this step was not done prior to the classifications being conducted. Therefore, a change analysis would have been a difficult step to fit in, as it requires that the images be correctly referenced to ground points and, more importantly, to each other.
The results of the classifications and visual comparison of the two images showed some important differences in the land cover data between the two years. Under close inspection, there is a slight decrease in the size of Lake Victoria, as can be seen in the receding shoreline in the Asembo region. Additionally, there is a large change from agricultural land and pasture to bare soil between 1984 and 2008. The forest and shrub areas fluctuate slightly between the two years, and appear to actually increase in area during the twenty four year time period.
Figure 3 - Classified map of the South Asembo region in 1984
Figure 4 - Classified image of South Asembo region in 2008
Figures three and four show the resulting classified images, as well as the histograms, names, and legends for the classes.
Overall, the results of this project were different than originally anticipated. While I did expect to see a recession of the Lake Victoria shoreline, I also expected to see more concrete visual results from deforestation and conversion to farmland in the region. Instead, the farmland and pasture areas actually decreased in size between 1984 and 2008. This could be due to differences in the month of the photo, however this problem was difficult to avoid if I wanted to obtain a satellite image with little cloud cover of the area.
There were many unforeseen issues that arose while completing this project that led to problems throughout the entire process. My partner for the project requested that we take an incomplete, which we did, but then chose to drop the class. As aa result, I had to complete the project on my own after considerable time had passed from when I took the course. Therefore, I was a little rusty on the operations in ERDAS Imagine, and had to redo large portions of the work. Additionally, there were some miscommunications about what steps had or had not been completed, which led to problems later on (such as thinking that the image rectification step had occurred when it had not).
Without these issues, the results of this project would likely have been more accurate. However, I still gained a great deal of knowledge from completing the process, particularly regarding the operations within ERDAS Imagine. Additionally, I learned valuable information about the problems that can occur when working in teams, and how foresight and planning ahead can help avoid these issues.