Flooding is a recurring disaster with severe consequences for human life, ecosystems, and infrastructure. Unfortunately, the prospective changes in climate induced by human activities will likely increase the intensity and likelihood of natural disasters. In low-income countries, the impact of flood events is exacerbated by the lack of proper meteorological data for effective flood modelling and forecasting. In the metropolitan city of Abidjan on the Ivory Coast, the last few decades have been marked by deadly flood events that have caused significant economic destruction. In this context, alternative data sources and valid modelling processes are essential for improved flood modelling and forecasting. This thesis aims to develop a hydrologic and hydraulic model for urban flood simulation in a poorly gauged catchment by examining various rainfall data sources and different modelling scenarios. The approach in the first stage combines satellite rainfall estimates (SRE) and ground rainfall data to evaluate the temporal distribution of extreme rainfall events (ERE) and predict extreme event scenarios through statistical and frequency analysis. Ground rainfall data from the National Meteorological Administration (SODEXAM) and four satellite real-time products (Tropical Rainfall Measuring Mission - TRMM, Climate Prediction Center (CPC) Morphing Technique - CMORPH, Precipitation Estimation from Remotely Sensed Information using Artificial Neural Network - PERSIANN, and Soil Moisture to Rain - SM2RAIN) were used. The pattern of extreme event distribution was assessed by counting the number of heavy, very heavy, and extreme rainfall events from 2007 to 2019 and noting the months in which they occurred. The maximum daily annual rainfall from the rain gauge data, spanning 1980 to 2019, was adjusted using log-normal, log-Pearson type 3, and Gumbel distribution functions (with the HYFRAN-PLUS program) to determine the best-fit distribution functions. Additionally, a 24-hour intensity-duration frequency analysis was performed using the Hydrognomon program. Through comparison with measurements, statistical extremes, and bias-corrected satellite data, extreme event scenarios were identified. The temporal distribution of ERE indicates that May and June are the months with the highest probability of ERE, accounting for 45% of past EREs. The Gumbel distribution fits best, as demonstrated by the Chi-square and Kolmogorov-Smirnov tests for ERE prediction in Abidjan. By comparing the bias-corrected satellite rainfall estimates with ground rainfall measurements from 1980 to 2019, several extreme rainfall events with return periods of up to 500 years were identified for further rainfall-runoff simulations. Secondly, the catchment hydrologic response to past flood events was analysed, and through various modelling scenarios and parameter investigations, a hydrological model was established and validated. The conceptually based, deterministic, and semi-distributed HEC-HMS model was used for computation. SCS-CN and Green-Ampt precipitation loss methods, Clark and SCS Unit Hydrographs direct runoff methods, and Muskingum and Kinematic-wave routing methods were combined in eight rainfall-runoff schemes. The changes in land use/cover and the impact of the recently built retention reservoirs on flood surface runoff were also assessed. Moreover, several bias correction methods were applied to SREs and implemented in the calibrated hydrologic model. Four selected satellite rainfall estimates (PERSIANN, PERSIANN Cloud Classification System - PERSIANN-CCS, and Integrated Multi-satellite Retrievals for Global Precipitation Measurement - IMERG version V7 and version V6) were compared to rain gauge data from 01/07/2015 to 30/12/2022 on a daily time scale. Then, three bias correction methods (Linear scaling – LS, Local intensity scaling – LOCI, and Power transformation – PT) were employed to correct SRE bias over the entire period and particularly during extreme rainfall event periods for the study area. The reliability of bias-corrected SREs in ERE evaluation and hydrological modelling was assessed. The hydrologic investigations show that the SCS-CN precipitation loss methods performed fairly well alongside the Clark unit hydrograph and Muskingum methods. The retention reservoirs led to reductions in peak discharge of 17% and 71% during the May and July 2018 flash flood events. Subsequently, land use maps from 2014, 2016, 2018, and 2020 were created using Landsat satellite image classification. The evaluation reveals that land use/cover underwent drastic changes from 2014 to 2020, with a relative intensification of 59.63% in built areas and reductions of 93.87% in forest and 69.93% in vegetation areas. This resulted in increases of 15.26% and 30.11% in peak flood and 31.19% and 30.79% in surface runoff for the May and July 2018 flash floods, respectively. The investigation of the reliability of SREs shows that all SREs exhibit low correlation at a daily time scale, with the best performance achieved by IMERG V7 (Correlation coefficient CC = 0.50) and IMERG V66 (CC = 0.46). It indicates that during EREs, bias correction methods could significantly enhance SREs; for example, a CC of 0.83 was achieved for IMERG V7 corrected with the LOCI method, while bias reductions were only slightly improved for the entire period due to rainfall intensity variability throughout the year, compounded by numerous drizzle days. The runoff simulation in HEC-HMS using PERSIANN corrected with the LOCI method achieved a Nash Sutcliffe Coefficient (NSE) of 0.61 for calibration and 0.71 for the validation event. Likewise, the LOCI bias correction method performed well for all simulations across the different SRE products. Thirdly, a plausible 2D hydrodynamic model using HEC-RAS 2D model was built through modelling component sensitivity analysis. The grid resolution, DEM manipulation, computational time step, infiltration models, and governing equations were investigated. The produced flood map was assessed using post-flood information from social media and newspapers and compared to a GIS-based flood hazard map. The bias-corrected SREs and flood retention reservoirs were further evaluated. The hydrodynamic modelling results indicated that most parameters and modelling components influenced the computational outcomes and should be analysed in detail when establishing a 2D hydrodynamic model, particularly for poorly gauged catchments. The June 2018 flood event was successfully reproduced with good agreement with social media and newspaper reports. The PERSIANN corrected LOCI satellite rainfall estimates produced a similar flood map to the ground rainfall, thus enhancing the potential of satellite rainfall estimates in flood modelling. The retention reservoirs significantly mitigated flood hazards by reducing flood velocity and depth by up to 50%, although the flood inundation area was decreased by only 6%. The GIS-based flood hazard map closely resembled the hydrodynamic map, with a slight increase in flood hazard depicted by the GIS-based approach. The study confirms the applicability of SREs for flood modelling across various scenarios for a given catchment. It also emphasises the significance of choosing appropriate bias correction methods and SREs for hydrological applications. Additional flood mitigation solutions, such as green infrastructure, could be implemented and analysed in the future. The results presented in this thesis provide a strong foundation for developing a robust flood modelling framework for a poorly gauged catchment. Using hydrologic and hydraulic models, the methodology demonstrated here can be replicated in other poorly gauged catchments to address flood issues.
Gogoua Habib Gogoua (Thu,) studied this question.