MODELLING AI TECHNOLOGIES TOWARDS PREDICTION OF DISASTERS RELATED TO CLIMATE CHANGE: CASE STUDY OF NORTH RIFT, KENYA

The study explores the application of artificial intelligence (AI) technologies for predicting climate change-induced disasters in Kenya’s North Rift region. The North Rift, characterized by diverse topography including highlands, valleys, and arid plains, has experienced increasing frequency and severity of climate-related disasters such as floods, droughts, and landslides over the past decade. These events have significantly impacted agricultural productivity, water resources, infrastructure, and community livelihoods.
The study employs machine learning algorithms, including random forests, convolutional neural networks, and long short-term memory (LSTM) networks, to analyze historical meteorological data, satellite imagery, and ground-based observations. This multi-modal approach enables the integration of traditional climate indicators with novel predictive features derived from remote sensing. The research leverages data from Kenya Meteorological Department stations, climate analysis products, and Earth observation satellites to develop regionally calibrated prediction models. Preliminary findings demonstrate that AI-based systems outperform conventional statistical methods in predicting the onset, intensity, and spatial distribution of climate disasters in the region. Notably, the LSTM models achieved 78% accuracy in forecasting drought conditions three months in advance, while CNN-based image analysis shows promising results in identifying flood-prone areas with 82% precision. The research addresses challenges related to data availability and quality through novel data fusion techniques and transfer learning approaches that adapt global climate models to local contexts. The study further examines the integration of AI predictions into existing early warning systems and disaster management frameworks. Stakeholder interviews with local government officials, community representatives, and disaster management agencies reveal both opportunities and barriers for effective implementation. Key recommendations include capacity building for local meteorological services, development of userfriendly prediction interfaces, and community-based participatory approaches for validation and refinement of AI outputs. This research contributes to the growing field of climate AI and demonstrates the potential of machine learning in enhancing disaster preparedness and resilience in vulnerable regions.The findings provide a foundation for developing scalable AI-based early warning systems that can be adapted to similar ecological contexts across East Africa