Modelling and forecasting of flood events for decision support in integrated water resources management in Gucha-Migori river basin, Kenya

Abstract

Floods are devastating natural disasters frequently striking many river basins such as Gucha-Migori River Basin nearly every year. However, non-structural countermeasures have not been fully explored and implemented to reduce damage, risk, and vulnerabilities of flood events. In light of this, the main objective of the study was to model and forecast flood events in the Gucha-Migori River Basin for decision support in integrated water resource management. Modelling the response of flood events to precipitation variability using the HEC-HMS model for the period (1973-2015) was the first activity conducted. Then a relationship between flood magnitude and frequency was developed using probability distribution model. Coefficient of Determination (R2), Goodness of Fit test, and Best-Fit Distribution Curve guided selection of the suitable distribution model. Finally, flood events and their respective annual exceedance probabilities were forecasted using Artificial Neural Networks (ANNs) and the Gumbel distribution model for the period (2015-2052). From the results, independent flood events were detected for the years 1969, 1971, 1974, 1977, 1969, 1981, 1982, 1983, 1985, 1990, 1996, 2004, 2006, 2007, 2010, 2011, and 2013. The R2 and Nash values for HEC-HMS model calibration and validation of daily river discharge at sub-basin 1 were 0.52 and 0.36, and, 0.42 and 0.31 respectively. The correlation coefficients between daily precipitations for JFM, AMJ, JAS, and OND, and the corresponding river discharges had a significant positive correlation at the significant level of 0.05. It was revealed that the probability of a flood event of a magnitude equal to or exceeding the lowest (240.8 m3/s) and the highest (423.90 m3/s) occurring for a particular were 0.98 and 0.02 respectively. Considering the predictions from Gumbel, Log P3, Log N, and Normal models, for the return periods 1.05 and 200 years, the flood magnitude estimated for G-M River Basin would be 221.38 and 466.65, 233 and 478, 178.19 and 367.38, and 209 and 402 m3/s. Gumbel distribution curve was selected to be the best fit for the Gucha-Migori River Basin. Using the comparative analysis, the network topology of 1-20-1 was adopted as the best NAR model for forecasting because of its minimum values of R for training (9.20e-1), validation (9.27e-1), and testing (9.15e-1). The maximum forecasted flood magnitudes were 336, 391, 433, 299, 389, 502, and 483 m³/s for the periods 2014–2018, 2019–2023, 2024–2028, 2029–2033, 2034–2038, 2039–2043, and 2048–2052, respectively. Their respective annual non-exceedance probabilities from the Gumbel’s curve and derived equation were 84.6, 96, 98, 64, 96, 99.6, and 99 %. The information from the research offers prerequisite flood response plans and provides early flood detection and forecasting for planning for the management of flood risks and for preparedness in the Gucha-Migori River Basin