Flood Flow Simulation Integrating Remote Sensing Techniques

Zotou, Ioanna

dc.contributor.author	Zotou, Ioanna
dc.date.accessioned	2025-09-22T09:46:28Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/62515
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.30211
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/3.0/gr/	*
dc.subject	flood modeling	en
dc.subject	hydrological modeling	en
dc.subject	remote sensing	en
dc.subject	uncertainty analysis	en
dc.subject	calibration framework	en
dc.subject	SAR imagery	en
dc.subject	Προσομοίωση πλημμύρας	el
dc.subject	Υδρολογική μοντελοποίηση	el
dc.subject	τηλεπισκόπηση	el
dc.subject	ανάλυση αβεβαιότητας	el
dc.subject	πλαίσιο βαθμονόμησης	el
dc.subject	απεικόνιση SAR	el
dc.title	Flood Flow Simulation Integrating Remote Sensing Techniques	en
dc.contributor.department	Laboratory of Reclamation Works and Water Resources Management	el
heal.type	doctoralThesis
heal.secondaryTitle	Προσομοίωση Πλημμυρικής Ροής με Χρήση Τεχνικών Τηλεπισκόπησης	el
heal.classification	Hydrology/Hydraulic engineering	en
heal.dateAvailable	2026-09-21T21:00:00Z
heal.language	en
heal.access	embargo
heal.recordProvider	ntua	el
heal.publicationDate	2025-07-16
heal.abstract	Floods rank among the most frequent and hazardous natural disasters, posing significant threats to human life, as well as the natural and built environment. The detrimental impacts of floods are anticipated to be further compounded in the future, driven by both climate change -which tends to enhance floods frequency and magnitude- and a range of additional factors, including the increase in population density, fast economic growth, urbanization and uncontrolled development along the floodplains, which collectively tend to enhance the vulnerability and risk in flood-prone areas. The above situation emphasizes the necessity for developing appropriate mitigation strategies for managing flood risk and alleviating the subsequent damage. Within this context, flood forecasting and early warning systems have attracted substantial interest, driven, among others, by the recent advancements in computing power and technology. The development of such systems typically relies on the coupling of a meteorological with a hydrological and/or a hydrodynamic component, with the latter being integrated by means of either sophisticated physics-based models or statistical data-driven techniques. Although computationally intensive, detailed physics-based simulators are generally accepted as more accurate in the representation of the flood phenomenon and they, thus, constitute a frequently used technique for flood forecasting and inundation mapping. Despite the increasing interest in flood modeling, hydrodynamic models continue to be subject to significant uncertainties arising from multiple stages of the modeling process. One of the most debated aspects is the choice of the appropriate model, especially in terms of the dimensionality (1D, 2D or coupled 1D/2D), which remains a contentious and critical decision influencing the accuracy of the simulation results. Beyond the well-documented uncertainties inherent in the modeling procedure, the intrinsically complex nature of floods combined with the substantial computational demands and the complete absence or limited availability of hydrometric data required for model calibration, poses further challenges for performing accurate and efficient flood modeling at high resolutions. In particular, the inadequacy of in-situ observations to independently support calibration and validation procedures has increasingly prompted a shift toward the use of satellite remote sensing products, which in turn enable systematic and large-scale flood mapping and monitoring at high resolutions. As a result, the incorporation of remote sensing techniques in flood modeling has attracted much attention lately, while the contribution of remotely sensed-derived products, especially in model calibration, has been primarily explored by considering either the flood extent or the indirectly retrieved water depth data. In light of the above, the present dissertation seeks to introduce a comprehensive investigation into the integration of satellite remote sensing with hydrodynamic modeling, specifically tailored to the particular hydrologic and physiographic characteristics of Greek riverine systems. Within this context, it ultimately aims to establish a robust framework for supporting decision-making in modeling-related issues and formulate an efficient strategy for the use of complex hydrodynamic models in ungauged, or insufficiently gauged basins. Emphasis is, thus, placed on the integration of remote sensing products, assessing the capabilities and limitations of different kinds of SAR-based data and techniques, to enhance the accuracy and reliability of hydrodynamic modeling. According to the above, the present thesis is organized around three primary objectives, which are: (a) to assess the overall applicability and inherent limitations of different remote sensing techniques and data in supporting flood modeling applications small- to medium-sized riverine systems, which are representative of the hydro-morphological characteristics of the Greek territory; (b) to conduct a thorough investigation into the factors contributing to uncertainty in hydrodynamic modeling outcomes, seeking to establish a practical guide for assisting modelers in making informed decisions and addressing common challenges, among which the selection of the appropriate modeling approach; and (c) to establish and implement a holistic framework for the calibration and validation of 2D hydraulic-hydrodynamic models, in complex and ungauged or insufficiently gauged basins, leveraging as benchmarks diverse satellite-derived products alongside efficient optimization techniques. To meet the above-described objectives, this research explores the freely available Sentinel-1 SAR products and evaluate their potential—in terms of accuracy, spatial and temporal resolution—for supporting flood modeling, in two distinct river catchments in Central Greece, namely the Spercheios and Peneios River Basins. In addition to the data itself, the influence of the selected SAR processing technique on the resulting outcomes is explored. To this end, two flood delineation approaches, namely a simplified single-image thresholding method and the Flood Mapping Python toolbox (FLOMPY), an automated approach based on SAR statistical temporal analysis, are used and comparatively tested to obtain insights on their relative appropriateness for supporting flood modeling applications. Finally, the comparative effectiveness of different satellite-derived end products, namely the flood inundation extent and the indirectly retrieved flood water depth, is evaluated within the scope of model calibration and validation. Uncertainty in hydrodynamic modeling was investigated in both Peneios and Spercheios River cases, offering, in this way, important insights into the applicability of the proposed framework, and allowing the identification of similarities or potential inconsistencies in the derived conclusions, due to varying flood conditions, complexity of the modeled systems and topographic characteristics. This phase of the research, thus, initiated with a Local Sensitivity Analysis (LSA), conducted for a certain river reach and flood event, occurred in late February 2018 in Peneios River, and aimed at providing preliminary insights on how different uncertainty sources impact on the predictive performance of a relatively simple 1D hydrodynamic model. A limited set of uncertain factors and configurations were investigated, compared to the analysis undertaken in the Spercheios River, including: (a) inflow discharge; (b) the variation in roughness coefficient; and (c) the spatial resolution of the terrain data. Model reliability in replicating the flood extent for the reconstructed event was assessed leveraging as a benchmark the flood extent extracted through SAR image processing, while model performance was quantified using a variety of widely accepted area-based measures of fit, i.e., the Critical Success Index (CSI), Success Index (SI), Hit Rate (H), False Alarm ratio (F), Accuracy (Acc) and Error Bias (E). Subsequently, the analysis was extended to the Spercheios River, where a comprehensive investigation of: (a) structural; (b) parametric; and (c) performance criteria-induced uncertainties was conducted, placing emphasis on interactions among the above uncertain components, as well. Within this context, three distinct hydrodynamic models, namely a coupled 1D/2D, a fully 2D and a coarse-resolution 2D model, were developed and their comparative response was assessed under the variation in five model variables, namely the: (a) roughness coefficient within the main channel and (b) across the two dominant land cover types in the floodplains; and (c) the variation in upstream discharge. Overall, the uncertainty investigation undertaken in Spercheios River was aimed at addressing the following issues: (a) investigate the impact of model structure -represented through the model dimensionality and simplifications in modeled system geometry- in the model output for identical input data and boundary conditions; (b) identify the most influential model variables and explore whether this is altered depending on the dimensional approach and performance metric; (c) examine the sensitivity of the individual performance metrics, with respect to the dimensional approach; (d) identify the optimum/worst dimensional approach according to each index; and (e) attempt to classify the utilized performance metrics in terms of how they rank the different modeling approaches. The overall framework implemented in the specific phase of the research was based upon a coupled hydrologic-hydrodynamic approach and involved the following five distinct parts: (a) the generation of the hydrologic forcing for use in the hydrodynamic simulation, coupling hydrologic modeling with a Monte Carlo Uncertainty Analysis; (b) the development of the three individual hydrodynamic model geometries, each representing a distinct dimensional approach; (c) the implementation of a LSA to test the influence of selected model parameters and upstream inflow, on the response of all three models, investigating each time exactly the same parameter combinations for all geometries tested; (d) the quantification of model response with respect to the derived flood extent for each model configuration scenario, considering a variety of performance metrics and using as benchmark the satellite-derived flood map; and (e) the sensitivity quantification and final evaluation. The final stage of the present dissertation, being also the overarching goal of the research, concerns the integration of the insights and lessons learned from the two preceding phases with the view to formulate a comprehensive strategy for the efficient calibration and validation of 2D hydraulic-hydrodynamic models, in complex and ungauged or insufficiently gauged basins, leveraging different satellite-derived data alongside efficient optimization techniques. The proposed framework is developed and tested in Spercheios River Basin, while also demonstrating the potential for adaptation to other regions. This stage is based on a coupled hydrological-hydrodynamic approach, and comprises three distinct components, namely: (a) the pre-processing or pre-calibration phase, (b) the calibration phase and (c) the validation phase. The pre-processing phase is further subdivided into two separate steps. The first step focuses on eliminating uncertainties in the hydrodynamic model results induced by factors unrelated to the parameters intended for calibration. Such factors were deemed to be the inflow boundary conditions, the accuracy of the terrain data, the absence of critical flow control structures and other modifications necessary for improving the fidelity of the modeled system. Within this context, a calibration of the upstream catchment runoff was undertaken, whereas a refined version of the 2D hydrodynamic model was built. This refined model incorporated an update of the underlying topography at key locations, the inclusion of key structural elements such as culverts, bridges etc., and additional adjustments aimed at increasing detail in model representation, with the purpose to address the so-called “leaking effect” and, consequently, enhance the reliability of the simulated outputs. The second step of the pre-calibration phase relates to the selection of the calibration parameters. To address the high computational demands typically associated with the screening procedure in complex models, a hybrid approach is proposed herein, which integrates a “manual” pre-screening phase, dedicated to perform a prior evaluation and exclusion of parameters anticipated to have negligible impact on model outcomes, with a Global Sensitivity Analysis (GSA) to prioritize the most influential among the parameters which passed the pre-screening phase. The suggested framework is complemented with the calibration-validation phase, which involves the use of a brute-force grid-search optimization technique tailored to the high computational demands of the detailed hydrodynamic models. The latter is a simple and cost-effective technique, which considers interactions between model parameters, yet being easily adjustable to the available computational budget, by properly tailoring the extent and step size of the parametric space. Therefore, the overall framework adheres to the principles of model parsimony and parametric abstraction, in order to reduce computational cost and mitigate equifinality issues. Two independent real-world flood events and the respective hydrological and Sentinel-1 SAR data were employed for calibrating and validating the refined 2D hydrodynamic model. During both the GSA and calibration stages, a variety of performance measures were exploited to gain a comprehensive insight on model response at several levels, while also assessing the comparative suitability of the examined metrics for supporting the calibration procedure. Model calibration was initially performed considering as benchmark the flood inundation extent, as derived from the FLOMPY algorithm. Subsequently, the Floodwater Depth Estimation Tool (FwDET) version 2.1 was employed to derive the spatial distribution of flood water depths across the model domain and explore the potential for better constraining the hydrodynamic model with respect to water depth prediction, as well. The latter served both testing the validity of the indirectly retrieved water depth product and evaluating its comparative utility for supporting calibration in relation to the use of flood extent data. Overall, the outcomes of this research revealed considerable inter-dependencies among the components contributing to uncertainty in hydrodynamic modeling, highlighting the complexities inherent in real-world modeling applications. With regard especially to the analysis implemented in the Spercheios River, the results revealed a tight nexus between the selected modeling approach, the performance criteria considered and the derived outcomes, in a way that impeded a unique demonstration of the most influential variables. In that sense, the demonstration of the most sensitive parameters was found to be influenced by both the selected modeling approach and performance metric considered, implying that the above factors should be accounted for by the modelers when a sensitivity analysis is to be performed. Moreover, model structure was found to substantially affect the simulation outcomes and the associated performance, even for identical input data and boundary conditions, whereas the optimum set of model parameter values proved to differ depending on both the adopted modeling approach and metric considered, thus emphasizing the importance of context-specific calibration. According to the results, the coarse-resolution or “simplified” 2D model consistently underperformed, while showing, in parallel, markedly lower sensitivity to the variation in model variables. This finding showcased the low potential of over-simplistic approaches for achieving an optimal model, particularly in cases of small rivers with complex topography, as the variation in parameter values and boundary conditions cannot fully compensate for the severe structural limitations. Effective modeling should be, thus, oriented towards an accurate representation of the area’s micro-topography, while preserving an optimal balance between model complexity, accuracy and computational cost. Furthermore, the incorporated metrics were found to differently rank the investigated modeling approaches as for their average performance. Hence, the analysis led to the formulation of two distinct groups of performance criteria; according to the 1st group which encompasses the CSI, H and SI, the fully 2D hydrodynamic model is ranked as ‘best’ outperforming the coupled 1D/2D approach, whereas according to the 2nd group (F, E and Acc) exactly the opposite occurs. The coarse-resolution 2D model emerges as the ‘worst’ approach in all cases. The above ranking also enabled the quantification of the anticipated percent improvement when passing from a ‘worse’ to a ‘better’ modeling approach, offering valuable implications for guiding model selection based upon accuracy requirements and computational sources. The LSA implemented in the Peneios River interestingly demonstrated certain inconsistencies with regard to the response of the different performance metrics, overall indicating a decrease in interdependencies among the various model components as the complexity of the modeled system decreases. Hence, in the above analysis, the selection of the performance metric was not found to affect the demonstration of the influential variables, whereas it proved to exert a less radical influence on the identification of the optimum parameter set. In addition, the comparative consideration of the results revealed apparent weaknesses of the investigated performance metrics, emphasizing that metrics encompassing areas correctly predicted as non-flooded by the model, may introduce substantial bias in performance evaluation results, particularly for model domains disproportionally large relative to the actual flood extent. The overarching goal of the present thesis has been the derivation of a robust framework for the efficient calibration and validation of complex hydrodynamic models, leveraging satellite data and efficient optimization techniques. In this regard, the proposed strategy for calibrating the parameters of a 2D hydrodynamic model using a brute-force grid search optimization technique was found to be effective, enabling the identification of a near-optimal solution with reduced computational demands. In addition, the hybrid screening procedure adopted, which comprised a manual combined with a GSA-based phase, demonstrated sufficiency in accordance with the principles of model parsimony and parametric abstraction, effectively reducing the computational cost and equifinality issues. The calibration results indicated substantial disagreements among the various performance metrics stressing that particular caution should be exercised in the selection of the objective function. CSI and SI indices proved to be more effective in the calibration context, being capable of reflecting the optimum model response and offering a comprehensive view of the overall performance, balancing all aspects. Conversely, Error Bias and Accuracy indices were found to be heavily biased towards avoiding overprediction, thus failing to reliably demonstrate optimum model performance. Although the RMSE is considered a rather detailed metric offering a cell-by-cell flood depth evaluation, the identification of its relative efficiency was hampered by the limitations of FwDET in accurately estimating water depths. Specifically, the algorithm’s reliability was found to decrease as the extent of inundated areas increased, while being heavily affected by the micro-topography of the area. Overall, the calibration results indicated equifinality issues stemming from the complex and non-linear nature of the model and the spatial interrelationships among the model variables, whereas GSA results were found to be particularly valuable in identifying potential equifinality challenges early in the process. With respect to the remote sensing techniques used in this thesis, although the manual flood delineation approach offered quick and decent insights, it proved to clearly underestimate flood extent due to the loss of spatial detail resulting from spatial filtering. In contrast, the FLOMPY approach, leveraging multi-temporal data, produced rather coherent inundation extent outcomes, more aligned with hydrodynamic simulations, thus, emphasizing the valuable role of multitemporal information for surmounting classification errors. As a whole, despite the overall valuable role of SAR data in enhancing model reliability and guiding modeling procedure, Sentinel-1 C-band data were found to pose serious challenges in dense vegetation and narrow river channels due to resolution and revisit frequency constraints.	en
heal.sponsor	This Doctoral Dissertation has been funded by the Special Account for Research Funding (E.L.K.E.) of the National Technical University of Athens between December 2018 – December 2022.	en
heal.advisorName	Tsihrintzis, Vassilios A
heal.committeeMemberName	Tsihrintzis, Vassilios A
heal.committeeMemberName	Karathanassi, Vassilia
heal.committeeMemberName	Nalbantis, Ioannis
heal.committeeMemberName	Vangelis, Harris
heal.committeeMemberName	Baltas, Evangelos
heal.committeeMemberName	Dimitriou, Elias
heal.committeeMemberName	Karatzas, George
heal.academicPublisher	Σχολή Αγρονόμων και Τοπογράφων Μηχανικών	el
heal.academicPublisherID	ntua
heal.numberOfPages	254
heal.fullTextAvailability	false