Back to search

University of the Aegean

Development of a framework of hydrological models, machine learning algorithms and participatory approaches in river floods risk management in the context of climate change

Abstract

dc:description

The intensification of natural hazards, particularly the occurrence of sudden and rapid flash floods in urban catchments, has underscored the imperative for robust, integrated flood management strategies. The formulation of flood mitigation and adaptation plans are confronted with mounting complexities attributed to the uncertainties arising from climate change. As climate shifts influence hydrological dynamics in urban areas, the assessment and prediction of river basin hydrological patterns through models take on paramount importance, considering prevailing and projected climatic scenarios. Concurrently, comprehensive appraisal of adaptation measures and the exploration of alternatives such as Nature-based Solutions (NBS) and advanced early forecasting and warning systems are gaining prominence, reflecting a broader response to environmental shifts. These innovative approaches, encompassing climate regulation, stand as effective alternatives to conventional flood management practices, effectively reducing flood risk and fostering long-term resilience amid changing climate conditions. This thesis is dedicated to the exploration of flood risk management, with a specific focus on flash floods occurring within small urban basins experiencing intermittent flow, all within the overarching context of climate change. The methodological framework, underpinning the research objectives, encloses three objectives: i) A comprehensive assessment of climate change impact on the hydrological regime of an intermittent flow basin draining urban areas using an ensemble model of diverse climate change scenarios, ii)The integration of participatory methodologies targeting the selection and evaluation of flood mitigation strategies. Nature based solutions are examined regarding their efficiency to decrease the flood peak and their implementation cost, iii) The development of an early flood forecasting and warning system, fortified by the synergistic interplay of real time gauging station, a physical hydrological model, and a state-of-the-art deep learning algorithm. These objectives are analyzed in detail across chapters 2 to 4, aiming to address the interrelationship between river flow regime, climate change, and urban dynamics. Chapter 2 constitutes in the exploration of the hydrological regime variation of an intermittent flow river in the face of climate change modelling uncertainties. This chapter's central focus lies in studying variations on the hydrological response of an ephemeral Mediterranean stream, Kalloni river. The downscaling of large-scale climate projections sourced from five General Circulation Models (GCMs) is undertaken. These projections encompass three distinct emission scenarios (RCP26, RCP45, and RCP85) and undergo downscaling through the utilization of the Long Ashton Research Station - Weather Generator (LARS-WG) model. In total, this study examines eleven sets of GCM-RCP scenarios, thus capturing a wide range of future weather conditions. The temporal window of this investigation spanned from 2021 to 2080, with the projected climate data to be used as input time series into the calibrated version of the Kalloni river hydrological simulation with the Hydrologic Engineering Center's HEC-HMS model. Within this methodological framework, a series of annual and seasonal changes in runoff patterns are derived, illuminating the high uncertainty of climate change impact on hydrological dynamics within small intermittent flow drainage systems. In chapter 3 a new methodology is described, which integrates participatory approaches into the urban flood risk mitigating process. This chapter highlights the interdisciplinary nature of this endeavor and suggests the use of local and indigenous focus groups in participatory decision-making actions. Specifically, it adapts and implements a stakeholder engagement approach, which is realized through a series of workshops and interviews targeting various stakeholder groups. The primary objective of these activities is to select urban flood mitigation and adaptation nature-based solutions and establish a comprehensive framework for their planning and implementation. A part of the methodology is based on an adaptation of the "Vulnerability and Capacity Assessment (VCA)" framework developed by the Red Cross, utilizing specialized social research tools. Within this approach, diverse groups of stakeholders (locals, experts, and decision-makers) collaboratively evaluate specific Nature-based Solutions, prioritizing them based on selected sustainability indicators spanning social, economic, and environmental dimensions. The concluding phase involves designing and estimating the costs of favored scenarios, employing numerical models. Ultimately, this holistic process results in the development of a widely accepted and sustainable nature-based flood management alternative enhancing long-term urban resilience. Chapter 4 introduces a soft, non-structural adaptation approach, centered on developing an Early Flood Monitoring and Forecasting System (EMFS) with the primary objective of delivering timely and precise predictions of critical overflow levels in a small-scale urban stream. Intermittent flow basins, characterized by intense, short-duration storms, frequently experience flash floods, amplifying the need for a responsive early forecasting and warning system. This chapter tackles the challenge of programming a real-time system capable of efficiently detecting coming flash floods and responding rapidly to the demands of such scenarios. The EMFS is fortified by a network of telemetric river monitoring stations, a physical hydrological model, and an advanced deep learning algorithm, collaborating to predict critical water levels. Specifically, five distinct Long Short-Term Memory (LSTM) neural network architectures are employed to analyze time series data and make multi-step predictions for short-term lead times of 15 minutes, 30 minutes, 45 minutes, and 1 hour. Additionally, feature importance analysis and hyperparameter tuning techniques are applied to enhance flood forecasting accuracy. Eventually, this approach has demonstrated its reliability, representing a significant tool to fortify the area's resilience against flash floods. These chapters collectively provide a comprehensive overview of both challenges and solutions in urban flood management. By interpreting the hydrological implications of climate change, crafting sustainable nature-based solutions, and implementing cutting-edge forecasting systems with deep learning techniques, this research aims to enhance the resilience of urban areas to floods, especially in the context of an unpredictable climate future.

Degree

thesis:*
Grantor dc:publisher
University of the Aegean
Year dc:date
2023

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Koutsovili, Eleni-Ioanna
  • Κουτσοβίλη, Ελένη-Ιωάννα

Subjects

dc:subject × 27

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/54608

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Koutsovili, Eleni-Ioanna; Κουτσοβίλη, Ελένη-Ιωάννα. Development of a framework of hydrological models, machine learning algorithms and participatory approaches in river floods risk management in the context of climate change. University of the Aegean, 2023. http://hdl.handle.net/10442/hedi/54608