C. Ntigkakis, Reverse analysis and uncertainty assessment of major flood events under limited data availability: The case of Western Attica, November 2017, Diploma thesis, 111 pages, Department of Water Resources and Environmental Engineering – National Technical University of Athens, Athens, October 2018.
From November 14th until November 15th, 2017, a storm event of substantial, yet unknown, local intensity, has caused a flash flood in Western Attica, Greece. The flood was responsible for significant economic losses, mainly focused in the city of Mandra, as well as for 24 fatalities. Right after the incident, a debate arose about whether the devastating results were due to the extreme nature of the storm, or due to the poor flood protection works. In this research, we analyzed all available information sources in an attempt to reproduce the actual storm event and provide estimations about its magnitude, temporal evolution and return period. The primary data source was the observed point rainfall at three meteorological stations in the wider area around the city of Mandra. However, one could easily conclude that these local rainfall events were not significant enough to cause such a severe flooding. This realization was further supported by the indicative rainfall estimations provided by an X-band meteorological radar, which recorded an unusual storm pattern of very high intensity over a very limited area, that strongly affected a relatively small spatial extent upstream of Mandra. Nevertheless, neither the point observations not the highly uncertain radar data were sufficient for providing quantitative estimations about the extreme rainfall event. The most valuable information was found in the neighboring catchment of Sarantapotamos, which is equipped with automatic stage recorder that controls a drainage area of 144.6 km2 (hydrometric station at Gyra Stefanis). The available data sources were: (a) point rainfall data at a remote (upstream) meteorological station (Vilia); (b) 15-min stage data that allowed reproducing part of the rising limb of the flood hydrograph at Gyra Stefanis, just before the flood destroyed the instruments assembly; and (c) audiovisual material at the station area, providing valuable information about the temporal evolution of the flood. The aforementioned information, quantitative and qualitative, was used in an attempt to estimate the rainfall over the basin of Sarantapotamos through a reverse rainfall-runoff modelling approach. In this respect, we tested several parsimonious versions of lumped event-based schemes, and calibrated their input against the observed flows at Gyra Stefanis. As input we considered the areal rainfall over Sarantapotamos, embedding the observed hyetograph at Vilia and an unknown hyetograph at a hypothetical station. The proposed modeling schemes used the SCS-CN method to estimate the effective rainfall and two alternative approaches, i.e. a lag-and-route method and a parametric synthetic unit hydrograph to propagate the runoff to the basin outlet. All model versions contained one free parameter, i.e. the initial abstraction ratio, and an unknown initial condition, expressed in terms of a dimensionless coefficient describing the antecedent moisture conditions of the SCS-CN method in continuous mode. Initially, we made scenario-based investigations, which revealed the major uncertainty induced by the lack of information and the sensitivity of results against the arbitrary assignment of hypothetical values to the two model parameters. In order to better assess this uncertainty, we next employed Monte Carlo simulations by generating 1000 random sets of the model parameters from suitable distributions and next solving the calibration problem. Within the selection of distributions, for parameter sampling, as well as within calibrations, we took advantage of all available information about the hydrological regime of the basin, the soil conditions the days before the event, and the temporal evolution of the flood after the destruction of the stage recorder. Based on the outcomes of Monte Carlo simulations, we provided probabilistic estimations of the quantities of interest, i.e. the total rainfall over the study area, its temporal evolution, and the peak flow of Sarantapotamos at Gyra Stefanis. We also employed risk evaluations, by estimating the maximum intensities and associated return periods of the storm event across several time scales. These results were then compared to additional sources of information, in an attempt to justify the plausibility of the aforementioned analysis.