Abstract/Summary

The frequency of floods is rising and constitutes one of the main causes of detrimental consequences arising from natural disasters, not only in Kenya, but across the globe. The anticipation, forecasting and accumulation of science-based evidence of floods is a vital component in managing, preparing for and mitigating the impacts of severe events, from local to international scales. This research aims to explore ways to improve flood modelling and forecasting at the national scale. To achieve this, a multi-stage approach is adopted, first, to understanding the key aspects to consider when selecting an appropriate model for flood forecasting and modelling, secondly, to understanding the application of global reanalysis precipitation datasets to hydrological modelling as potential alternatives due to data scarcity challenges, and thirdly, to undertaking an analysis of trend detection in floods and possible shifts in flood timing. Five criteria are applied to a hydrological model selection framework following a filter sequence; an evaluation of twelve potential models is performed, and four potential model candidates are selected for flood applications for a Kenyan national forecasting centre. Model selection has shown that not all models are good at capturing and/or representing the important processes relevant to flood generation and a single model would not be applicable to the entire country, due to stark differences in the hydroclimatic characteristics of catchments, and model developments and upgrades should allow incorporation of such differing characteristics. Four reanalysis precipitation datasets are evaluated for their ability to be used for hydrological modelling. The choice of precipitation input is found to be the dominant component of the hydrometeorological modelling chain, creating the need to aggregate both sensitivity indices and performance statistics. Improvements have arisen from the introduction of ERA5 as a source of meteorological data. Performance varies by season and catchment, with wetland catchments obtaining relatively better scores compared to those in the semi-arid regions. Examination of trends in river flow series identified statistically increasing trends in annual floods for stations in proximity to each other, which is evidence of a spatially coherent pattern. and increasing flood frequency across Kenyan catchments, with observations showing a shift in timing and variability in flood occurrences in most parts of the country. This research has explored and provided an enhanced understanding of the avenues of improving flood modelling and forecasting in Kenya in terms of models, data, and historical flooding trends as well as seasonality and shifts in flood timing, which can inform future developments and operational flood forecasting for the end-users of an early warning system that can help mitigate the effects of floods in data-scarce regions such as Kenya.