Rainfall runoff modeling in Yala River basin of Western Kenya

Abstract

When rainfall is received on a watershed, depending on the ini al soil moisture content some of the water seeps underground while the excess forms surface water response. The nature of the runoff and its effects in the watershed can be represented by the applica on of hydrologic models to predict streamfl ow. In this study, the Geological Streamfl ow Model (GeoSFM) and the Muskingum Cunge (M-C) model were used to model the hydrologic processes of the Yala river network. The objec ve of the study was to develop a flood early warning system to mi gate poten al fl ood hazard risk exposed to the downstream inhabitants. Historical hydro-metric datasets of 1975-2005 were used for calibra on, verifi ca on and streamfl ow rou ng based on a split record analysis. For the runoff genera on, rainfall and evapora on datasets were provided by the Kenya Meteorological Department (KMD) while for model calibra on and verifi ca on, streamfl ow was obtained from Water Resources Management Authority (WRMA). To determine the hydrologic connec vity, the 30 meters by 30 meters Digital Eleva on Model was obtained from the Interna onal Centre for Research in Agro-Forestry (ICRAF). The Digital Soil Map of the World developed by Food and Agricultural Organiza on (FAO) and the Global Land Cover data of the United States Geological Survey (USGS) were used for model pa- rameteriza on. The soil moisture accoun ng and rou ng method transferred water through the subsurface, overland and river phases. The percentage of the square of the correla on coeffi cient (R2% value) was used to determine model performance. The GeoSFM modeled streamfl ow at the Bondo streamflow gauging sta on, coded 1FG02 where during the calibra on and verifi ca on phases, streamfl ow was modeled at R2 value of 80.6% and 87.3% respec vely. The M-C model routed streamfl ow from 1FG02 to the Kadenge streamflow gauging sta on, coded 1FG03 at R2 value of 90.8%, Muskingum K value of 2.76 hours and Muskingum X value of 0.4609. The error in predicted peak streamfl ow was 2.3% with a posi ve 1.5% error in predicted speed. This ensured a forecast of the me of peak streamfl ow on the safe side before the actual fl ood peak arrival at 1FG03 sta on. It was concluded that the GeoSFM and M-C models were hence useful tools for flood mi ga- on by issuing fl ood early warning messages defi ned by peak streamfl ow and fl ood wave travel me.