A Time Series Model For Forecasting Lake Expansion: Case Study Of Lake Baringo

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.