| dc.description.abstract | A flood is a natural disaster that refers to the temporal overflow of water on top of land that
was previously not inhabited by water. It can be caused by too much precipitation or even
outbursts of water reservoirs due to other reasons. Severe Floods have occurred in the Rift
Valley lakes since 2011 due to lake expansion. Floods in the Lake Baringo area have occurred
due to overflows of the lake and is a dangerous disaster leading to many pros rather than cons.
It is due to the major problems experienced that the need for the use of Machine Learning, GIS,
and Remote Sensing arose to help in monitoring, and creation of a forecast model to help create
awareness of the area that is likely to be affected by floods in the future years. The research
was guided by three objectives: Determining factors leading to the expansion of Lake Baringo,
mapping spatial-temporal change in the Lake Baringo region to help compare the changes,
comparing the time-series algorithms (LSTM and GRU) efficiency in the training of the dataset
and lastly developing a time-series model for forecasting the area growth of Lake Baringo.
Earlier researchers had used GIS and RS for the monitoring of similar cases but the element of
prediction was not well looked into. Machine Learning methods have also been used to create
prediction models but in the case of lake area expansion limited researchers had explored,
hence the identified gaps arose. The research design used was longitudinal and it comprised
two sets of data mainly satellite images and previously recorded data. Images were used for
classification to map the changes over time and to visualize the lake's growth, the other form
of dataset was used for analysis and creation of the model. GRU outperformed the LSTM
algorithm ass per metrics, it was found that Lake Baringo had expanded by 50% from the year
2011 mainly due to increased rainfall and reduced evaporation increasing the rate of
sedimentation which led to the rising of the lake level. The study was limited by the available
data and time used in the image analysis. The objectives were achieved and, in the future, better
models could be developed for numerous lakes in Kenya and not only Lake Baringo. | en_US |
