Votre recherche

RÉSUMÉ: «RÉSUMÉ: Les inondations sont reconnues comme l’une des catastrophes naturelles les plus fréquentes et destructrices à l’échelle mondiale. Leur gravité est exacerbée par les effets du changement climatique (augmentation des précipitations) et de la construction humaine (réduction de la capacité naturelle à absorber l’eau). Les structures construites dans des zones sujettes à l’eau, telles que les ponts et les barrages, sont généralement vulnérables aux événements d’inondation sévères. Pour les problèmes impliquant de l’eau fluide, les chercheurs en hydraulique supposent généralement que les structures sont "infinitement" rigides et utilisent des limites de paroi imperméables pour représenter les structures dans les modèles numériques. Cependant, les structures se déformeront, vibreront et pourraient même être endommagées lors d’un événement d’inondation sévère. Du point de vue d’un ingénieur structurel, il est important d’incorporer la flexibilité structurelle dans l’analyse de l’interaction fluide-structure (FSI). Étant donné que la taille du domaine fluide est significativement plus grande que celle des structures, un grand nombre d’éléments est généré, rendant l’analyse FSI chronophage, surtout pour les cas avec un canal 3D long et des maillages raffinés. Par conséquent, une méthode de modélisation simplifiée efficace et précise est nécessaire. De plus, le comportement hydrodynamique des structures telles que le pont dans un cours d’eau et la structure du barrage à l’extrémité d’un canal partiellement recouvert de glace n’est pas bien connu. Pour aborder ce problème, cette recherche a examiné numériquement les réponses structurelles avec l’impact de l’écoulement des inondations en tenant compte de la flexibilité structurelle, en se concentrant sur l’interaction dynamique entre l’eau fluide et les structures solides, les effets 3D des fluides et des structures, le glissement des structures (par exemple, le glissement du tablier du pont), et la présence d’une couverture de glace partielle positionnée au sommet de l’eau dans un canal.» ABSTRACT: «ABSTRACT: Floods are recognized as one of the most frequent and destructive natural disasters globally. Their severity is exacerbated by the effects of climate change (increased precipitation) and human construction (reduced natural capacity to absorb water). Structures built in waterprone areas, such as bridges and dams, are usually vulnerable to severe flood events. For problems involving fluid water, hydraulic researchers commonly assume that structures are "infinitely" rigid and use impervious wall boundaries to present the structures in numerical models. However, structures will deform, vibrate, and even be damaged during a severe flood event. From a structural engineer’s perspective, it is important to incorporate structural flexibility into the fluid-structure interaction (FSI) analysis. Because the size of the fluid domain is significantly larger than that of the structures, a large set of elements is generated, making the FSI analysis time-consuming, especially for cases with a long 3D channel and refined meshes. As a result, an efficient and accurate simplified modeling method is needed. Also, the hydrodynamic behavior of structures such as the bridge in a stream and the dam structure at the end of a partially ice-covered channel is not well known. To address this problem, this research numerically investigated the structural responses with the impact of flood flow considering the structural flexibility, focusing on the dynamic interaction between fluid water and solid structures, the 3D effects of fluid and structures, the sliding of structures (e.g. sliding of bridge deck), and the presence of partial ice cover positioned at the top of the water in a channel.»

QUIC is a modern transport layer internet protocol designed to be more efficient and secure than TCP. It has gained popularity quickly in recent years and has been adopted by a number of prominent tech companies. Its efficiency comes from its handshake design. The server and the client make both the transport layer acknowledgment and the TLS agreement during the same round trip. However this process makes the packets heavy and requires more processing on the server-side than TCP. This characteristic can be used as leverage by an attacker to compromise the computing resources of its victim. This thesis investigates the resilience of QUIC Protocol against handshake flood attacks and proposes a detection mechanism (QUICShield). I conducted comprehensive experiments to evaluate the resource consumptions of both the attacker and the target during incomplete handshake attacks, including CPU, memory, and bandwidth. We compared the results against TCP Syn Cookies under Syn flood attacks. The DDoS amplification factor was measured and analyzed based on the results. This work also proposes a detection mechanism based on a Bloom filter combined with Generalized Likelihood Ratio Cumulative Sum (GLR-CUSUM) to adapt to evolving attack patterns. It was implemented and deployed against real attacks to evaluate its efficiency. We showed that the QUIC Protocol design has a much larger DDoS amplification factor compared to the TCP, which means QUIC is more vulnerable to handshake DDoS attacks. However the mechanism proposed is accurate and efficient in terms of resources.

Abstract: In Canada, the annual runoff is predominantly influenced by snowmelt following the winter season, with a substantial portion (40-80\%) occurring during the spring period, leading to flooding in low-lying areas. Accurate prediction of streamflow is essential for hydropower production, effective flood management, necessitating the incorporation of comprehensive spatially distributed snow observations into hydrological models. This draws the attention to the research question " How can we utilize spatially distributed snow information at various spatial and temporal scales to enhance our understanding of snow processes and apply it for enhanced model calibration to improve hydrological model performance?" The first objective of this thesis is to investigate the utilization of spatially distributed snow information (SNODAS- SNOw Data Assimilation System) for the calibration of a hydrological model and to determine its impact on model performance. A distributed hydrological model, HYDROTEL, has been implemented in the Au Saumon River watershed using input data from ERA-5 Land for temperature data and MSWEP for precipitation data. Seven different calibration experiments are conducted, employing three different objective functions: Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), and the SPAtial EFficiency metric (SPAEF). These objective functions are utilized individually or in combination as part of multi-objective calibration processes. This study indicates that utilizing SPAEF for spatial calibration of snow parameters improved streamflow prediction compared to the conventional practice of using RMSE for calibration. SPAEF is further implied to be a more effective metric than RMSE for both sequential and multi-objective calibration. During validation, the calibration experiment incorporating multi-objective SPAEF exhibits enhanced performance in terms of NSE and KGE compared to calibration experiment solely based on NSE. The findings of this study hold significant relevance and potential applicability in emerging satellite technology, particularly the future Terrestrial Snow Mass Mission (TSMM). The study then explores the impact of temporal resolution and signal saturation for model calibration by using SNODAS data as proxy SWE observations mimicking the characteristics of the TSMM product to calibrate the HYDROTEL model. Despite the limitations of it's temporal resolution and signal saturation it is noteworthy that TSMM data exhibits significant potential for enhancing model performance thereby highlighting its utility for hydrological modeling. This study then focuses on the spatio-temporal analysis of snow processes influencing the spatial variability and distribution of snow depth in a small-scale experimental watershed. Drone photogrammetry is employed to capture spatially distributed snow information over the watershed during the winter seasons of 2022 and 2023. The photogrammetric data facilitated the generation of high-resolution digital surface models (DSMs). Empirical Orthogonal Function (EOF) analysis is applied to understand the spatial distribution of snow, enabling a detailed examination of various snow processes at the watershed scale. This thesis explores the added value of spatially distributed snow cover information in predicting spring runoff. Each part of the study contributes to a comprehensive understanding of the spatial distribution of snow and its significance in hydrology.

The coast is a complex environment that comprises seawater, underwater, soil, atmosphere, and other environmental factors. Traditional and new pollutants, represented by oil spills and microplastic (MPs), persist in posing a constant threat to the ecosystems and social-economic features of coastal regions. Besides, the shoreline is exposed to various environment conditions, which may significantly affect the behaviors of pollutants on beaches. An in-depth understanding of the occurrence and fate of pollutants in coastal areas is a prerequisite for the development of sound prevention and remediation strategies. Firstly, the physicochemical behavior of crude oil on various types of shorelines under different environmental conditions were reviewed. The penetration, remobilization, and retention of stranded oil on shorelines are affected by the beach topography and the natural environment. The attenuation and fate of oil on shorelines from laboratory and field experiments were discussed. In addition, the source, type, distribution, and factors of MPs in the coastal areas were summarized. What is more, the occurrence and environmental risk of emerging plastics waste—personal protective equipment (PPE)—in the coastal environment during and pandemic were discussed. Then, the role of natural nanobubbles (NBs) in the fate and transport of spilled oil were investigated through laboratory experiments and model simulations. NBs significantly increased the concentration of dissolved oxygen as well as changed the pH, zeta potential, and surface tension of the water. With the assistance of external energy, the bulk NBs enhanced the efficiency in oil detachment from the surface of the substrate. At the same time, the surface NBs on the substrate obstructed the downward transport of oil colloids. Considering the behavior between the NBs in two different phases and the oil droplets, the oil droplets tended to bind to the NBs. Next, the behavior and movement of various MPs in the presence of bulk NBs was explored. In the presence of NBs, the binding of MPs and NBs resulted in an increase in the measured average particle size and concentration. The velocity of motion of MPs driven by NBs varies under different salinity conditions. The increase in ionic strength reduced the energy barrier between particles and promoted their aggregation. Thus, the binding of NBs and MPs became more stable, which in turn affected the movement of MPs in the water. Polyethylene (PE1) with small particle size was mainly affected by Brownian motion and its rising was limited, therefore polyethylene (PE2) with large particle size rose faster than PE1 in suspension, especially in the presence of NBs. The effect of nanobubbles on the mobilization of MPs in shorelines subject to seawater infiltration was further studied. The motion of MPs under continuous and transient conditions, as well as the upward transport induced with flood were considered. Salinity altered the energy barriers between particles, which in turn affected the movement of MPs within the matrix. In addition, hydrophilic MPs were more likely to infiltrate within the substrate and had different movement patterns under both continuous and transient conditions. The motion of the MPs within the substrate varied with flow rate, and NBs limited the vertical movement of MPs in the tidal zone. It was also observed that NBs adsorbed readily onto substrates, altering the surface properties of substrates, particularly their ability to attach and detach from other substances. Finally, the changing characteristics and environmental behaviors of PPE wastes when exposed to the shoreline environment were examined. The transformation of chain structure and chemical composition of masks and gloves as well as the decreased mechanical strength after UV weathering were observed. In addition, the physical abrasion caused by sand further exacerbated the release of MPs and leachable hazardous contaminates from masks and gloves. In conclusion, the coastal zone is threatened by various pollutants, including traditional pollutants (like the oil spill) and emerging pollutants (like MPs). Due to the complexity of the coastal zone, the occurrence, transport and fate of pollutants can be controlled by many factors, and some factors that are ignored before can also alter the environmental behavior of pollutants in the coastal zone. Natural NBs can change the properties of the water environment and affect the surface properties of the substrate. Bulk NBs contribute to the oil detachment from the sand surface, and surface nanobubbles in the substrate obstruct the downward transport of oil colloids. The behavior and mobilization of MPs in the coastal `zone are subject to mutual forces between the substrate, MPs, NBs, and other factors. Coastal zones are not only the main receptor of pollutants from oceans and lands but also play a key role in their fate and transport.

Natural calamities like floods and droughts pose a significant threat to humanity, impacting millions of people each year and incurring substantial economic losses to society. In response to this challenge, this thesis focuses on developing advanced machine learning techniques to improve water height prediction accuracy that can aid municipalities in effective flood mitigation. The primary objective of this study is to evaluate an innovative architecture that leverages Long Short Term Networks - neural networks to predict water height accurately in three different environmental scenarios, i.e., frazil, droughts and floods due to snow spring melt. A distinguishing feature of our approach is the incorporation of meteorological forecast as an input parameter into the prediction model. By modeling the intricate relationships between water level data, historical meteorological data and meteorological forecasts, we seek to evaluate the impact of meteorological forecasts and if any inaccuracies could impact water-level prediction. We compare the outcomes obtained by incorporating next-hour, next-day and next-week meteorological data into our novel LSTM model. Our results indicate a comprehensive comparison of the usage of various parameters as input and our findings suggest that accurate weather forecasts are crucial in achieving reliable water height predictions. Additionally, this study focuses on the utilization of IoT sensor data in combination with ML models to enhance the effectiveness of flood prediction and management. We present an online machine learning approach that performs online training of the model using real-time data from IoT sensors. The integration of live sensor data provides a dynamic and adaptive system that demonstrates superior predictive capabilities compared to traditional static models. By adopting these advanced techniques, we can mitigate the adverse impacts of natural catastrophes and work towards building more resilient and disaster-resistant communities.

Cette thèse traite des aspects de la quantification de l'incertitude dans l'évaluation des ressources éoliennes avec les pratiques d'analyses d'incertitude et de sensibilité. Les objectifs de cette thèse sont d'examiner et d'évaluer la qualité des pratiques d'analyse de sensibilité dans l'évaluation des ressources éoliennes, de décourager l'utilisation d'une analyse de sensibilité à la fois, d'encourager l'utilisation d'une analyse de sensibilité globale à la place, d'introduire des méthodes d'autres domaines., et montrer comment les analyses d'incertitude et de sensibilité globale ajoutent de la valeur au processus d'aide à la décision. Cette thèse est organisée en quatre articles : I. Une revue des pratiques d'analyse de sensibilité dans l'évaluation des ressources éoliennes avec une étude de cas de comparaison d'analyses de sensibilité individuelles et globales du coût actualisé de l'énergie éolienne offshore ; II. Technique Quasi-Monte Carlo dans l'analyse de sensibilité globale dans l'évaluation des ressources éoliennes avec une étude de cas sur les Émirats Arabes Unis; III. Utilisation de la famille de distribution Halphen pour l'estimation de la vitesse moyenne du vent avec une étude de cas sur l'Est du Canada; IV. Étude d'évaluation des ressources éoliennes offshore du golfe Persique avec les données satellitaires QuikSCAT.Les articles I à III ont chacun donné lieu à une publication évaluée par des pairs, tandis que l'article IV - à une soumission. L'article I propose des classifications par variable de sortie d'analyse de sensibilité, méthode, application, pays et logiciel. L'article I met en évidence les lacunes de la littérature, fournit des preuves des pièges, conduisant à des résultats d'évaluation erronés et coûteux des ressources éoliennes. L'article II montre comment l'analyse de sensibilité globale offre une amélioration au moyen du quasi-Monte Carlo avec ses plans d'échantillonnage élaborés permettant une convergence plus rapide. L'article III introduit la famille de distribution Halphen pour l'évaluation des ressources éoliennes. Article IV utilise les données satellitaires SeaWinds/QuikSCAT pour l'évaluation des ressources éoliennes offshore du golfe Persique. Les principales contributions à l'état de l'art avec cette thèse suivent. À la connaissance de l'auteur, aucune revue de l'analyse de sensibilité dans l'évaluation des ressources éoliennes n'est actuellement disponible dans la littérature, l'article I en propose une. L'article II relie la modélisation mathématique et l'évaluation des ressources éoliennes en introduisant la technique de quasi-Monte Carlo dans l'évaluation des ressources éoliennes. L'article III présente la famille de distribution de Halphen, de l'analyse de la fréquence des crues à l'évaluation des ressources éoliennes. <br /><br />This dissertation deals with the aspects of quantifying uncertainty in wind resource assessment with the practices of uncertainty and sensitivity analyses. The objectives of this dissertation are to review and assess the quality of sensitivity analysis practices in wind resource assessment, to discourage the use of one-at-a-time sensitivity analysis, encourage the use of global sensitivity analysis instead, introduce methods from other fields, and showcase how uncertainty and global sensitivity analyses adds value to the decision support process. This dissertation is organized in four articles: I. Review article of 102 feasibility studies: a review of sensitivity analysis practices in wind resource assessment with a case study of comparison of one-at-a-time vs. global sensitivity analyses of the levelized cost of offshore wind energy; II. Research article: Quasi-Monte Carlo technique in global sensitivity analysis in wind resource assessment with a case study on United Arab Emirates; III. Research article: Use of the Halphen distribution family for mean wind speed estimation with a case study on Eastern Canada; IV. Application article: Offshore wind resource assessment study of the Persian Gulf with QuikSCAT satellite data. Articles I-III have each resulted in a peer-reviewed publication, while Article IV – in a submission. Article I offers classifications by sensitivity analysis output variable, method, application, country, and software. It reveals the lack of collective agreement on the definition of sensitivity analysis in the literature, the dominance of nonlinear models, the prevalence of one-at a-time sensitivity analysis method, while one-at-a-time method is only valid for linear models. Article I highlights gaps in the literature, provides evidence of the pitfalls, leading to costly erroneous wind resource assessment results. Article II shows how global sensitivity analysis offers improvement by means of the quasi-Monte Carlo with its elaborate sampling designs enabling faster convergence. Article III introduces the Halphen distribution family for the purpose of wind recourse assessment. Article IV uses SeaWinds/QuikSCAT satellite data for offshore wind resource assessment of the Persian Gulf. The main contributions to the state-of-the-art with this dissertation follow. To the best of author’s knowledge, no review of sensitivity analysis in wind resource assessment is currently available in the literature, Article I offers such. Article II bridges mathematical modelling and wind resource assessment by introducing quasi-Monte Carlo technique to wind resource assessment. Article III introduces the Halphen distribution family from flood frequency analysis to wind resource assessment.

Streamflow forecasting is important for managing water resources in sectors like agriculture, hydropower, drought management, and urban flood prevention planning. Our study examines short and long lead-times to create a framework for streamflow forecasting that can benefit water resource management and related sectors. To improve streamflow forecasts for up to ten days of lead-time, the study first focuses on improving initial conditions using an ensemble Kalman filter as a data assimilation method. The goal is to regulate the hyperparameters of the ensemble Kalman filter for each season to produce more accurate forecasts. A sensitivity analysis is conducted to identify the best hyperparameter sets for each season, including uncertainty in temperature, precipitation, observed streamflow, and the water content of three state variables - vadose zone, saturated zone, and snowpack - from the CEQUEAU model. Results indicate that improving initial conditions with the ensemble Kalman filter produces more skillful forecasts until a 6-day leadtime. Temperature uncertainty is particularly sensitive and varies across seasons. The vadose zone state variable was identified as the most important and sensitive state variable, and updating all state variables systematically may not be necessary for improving forecast skill. Recent machine learning advances are improving short-term streamflow forecasting. One such method is the Long Short-Term Memory (LSTM) model. In general, neural networks learn from regression as relationships exist between input-output. However, LSTM models have a feature named ‘forget gate’, which enables them to learn the relationship between inputs (e.g., temperature and precipitation) and output (streamflow), and also to capture temporal dependencies in the data. The study aimed to compare the performance of the Long ShortTerm Memory (LSTM) model with data assimilation-based and process-based hydrological models in short-term streamflow forecasting. All three models were tested using the same ensemble weather forecasts. The LSTM model demonstrated good performance in forecasting streamflow, with a Kling-Gupta efficiency (KGE) greater than 0.88 for 9 lead-times. The LSTM model did not incorporate data assimilation, but it benefited from observed streamflow until the last day before the forecast. This is because the LSTM model learned and incorporated knowledge from the previous days while issuing forecasts, similar to how data assimilation updates initial conditions. The study results also showed that the LSTM model had better performance up to day 6 of lead-time compared to the data assimilation-based models. However, training the LSTM model separately for each lead-time is a time-consuming process and is a disadvantage compared to the data assimilation-based methods. Nonetheless, the study demonstrated the potential of machine learning techniques in improving streamflow forecasting. The forecasting of streamflow for long lead-times such as a month usually involves the use of historical meteorological data to create probable future scenarios, as meteorological forecasts become unreliable beyond this lead-time. In this study, we proposed a novel method for streamflow forecasting based on ensemble streamflow forecasting (ESP) filtering, using a Genetic Algorithm (GA) to filter forecast scenarios. This method quantifies the potential of historical data for each basin. This potential could be utilized to enhance the accuracy of streamflow forecasts. We sorted the selected and unselected scenarios to find out the common features between them, but the results did not help distinguish between the two groups. Nonetheless, the GA method can be used as a benchmark for future studies to improve longterm streamflow forecasting. This method can also be used to compare different forecast methods based on the potential shown by the GA method for a specific size of ESP members. For instance, if a method uses large-scale climate signals to filter ESP members, the forecast skill result could be compared with the potential of historical data for that particular size of ESP members.

Reliable long-term streamflow forecast is essential in water resources management and plays a key role in reservoir management and hydropower generation. Properly framing the uncertainty is the key issue in providing a reliable long-term streamflow forecast, and probabilistic forecasts have been used to this effect. In a probabilistic approach, each observed historical data is taken as a possible realization of the future. Non stationarity of hydrometeorological variables, either due to the climate internal variability or anthropogenic change, is another important problem for long-term streamflow forecasts as it is becoming increasingly clearer that past historical data may not adequately represent the current climate. Therefore, there is a need to develop flexible approaches taking into account non-stationarity for long-term streamflow forecasts. Resampling past historical time series is the main approach used for probabilistic long term streamflow forecasts. However, non-stationarity is a key issue of resampling approaches. One possible approach is to make use of a stochastic weather generator coupled to a hydrological model to generate long-term probabilistic streamflow forecasts. Weather generators can easily be modified to account for climatic trends and therefore have the potential to take non-stationarity into account. However, before weather generators can be modified to account for climate non-stationarity, it is first necessary to evaluate whether the modeling chain consisting of a stochastic weather generator and a hydrological model can generate probabilistic streamflow forecasts with a performance similar to that of more traditional resampling approaches. The first objective of this study is therefore, to compare the performance of a stochastic weather generator against that of resampling historical meteorological time series in order to produce ensemble streamflow forecasts. Results indicate that while there are differences between both methods, they nevertheless largely both perform similarly, thus showing that weather generators can be used as substitutes to resampling the historical past. Based on these results, two approaches for taking non-stationarity into account have been proposed. Both approaches are based on a climate-based perturbation of the stochastic weather generator parameters. The first approach explored a simple perturbation method in which the entire length of the historical record is used to quantify internal variability, while a subset of recent years is used to characterize mean climatic values for precipitation, minimum and maximum temperatures. Results show that the approach systematically improves long-term streamflow forecasts accuracy, and that results are dependent on the time window used to estimate current mean climatic estimates. The second approach conditioned the parameters of a stochastic weather generator on largescale climate indices. In this approach, the most important climate indices are identified by looking at yearly correlations between a set of 40 indices and precipitation and temperature. A linear model is then constructed to identify precipitation and temperature anomalies which are then used to induce perturbations in the stochastic weather generator. Five different time windows are defined to determine the optimal linear model. Results show that temperatures are significantly correlated with large-scale climate indices, whereas precipitation is only weakly related to the same indices. The length of the time window has a considerable impact on the prediction ability of the linear models. The precipitation models based on short-duration time windows performed better than those based on longer windows, while the reverse was found for the temperature models. Results show that the proposed method improves long-term streamflow forecasting, particularly around the spring flood.

Dams are vital national assets that play a crucial role in water storage, hydroelectric power generation, and flood control. Globally, over 61,000 large dams have surpassed 50 years of service, and many show signs of deterioration. With over 300 dam failures recorded worldwide, the potential for catastrophic damage remains alarmingly high if these aging structures are not properly maintained and upgraded. Further, many of the existing dams were built upon outdated standards, and there is an increase in seismic hazards making it imperative to reevaluate their seismic performance to align with current safety standards. The need for improved dam safety measures is urgent, as dam owners, regulators, and policymakers grapple with the challenges of ensuring the structural integrity of aging dams in the face of growing risks. A key solution is shifting from traditional safety approaches to a modern, risk-based methodology, which addresses safety concerns more efficiently and economically. Various, global agencies have developed risk-based safety assessment guidelines; however, these often lack systematic implementation frameworks and sufficient reference studies, making them difficult for dam owners to adopt effectively. Furthermore, various uncertainties can impact the risk assessment and can complicate efforts to ensure dam safety. In this context, this research investigates uncertainties impacting seismic risk assessments for dams, including modeling choices, ground motion selection, aging, and material variability. Case studies of the Koyna Dam and Pine Flat Dam were used to evaluate these factors at each stage of performance evaluation: system response, fragility, and risk assessment. Key findings indicate that dam-foundation-reservoir (DFR) models incorporating acoustic elements exhibit less variability in system response, regardless of model complexity and solution procedure. Ground motion derived from the conditional mean spectrum (CMS) method yields better fragility estimates than the ASCE 7-16 standard, particularly for moderate to severe damage states. Additionally, aging and material variability significantly affect the dynamic characteristics of dams, with increased failure probabilities correlating with both age and return period. Based on these findings, the research proposes a comprehensive, systematic framework for risk-based seismic safety evaluation. This framework aligns with safety assessment objectives and ensures optimal use of computational resources.

Flooding, a major natural calamity, severely threatens communities and infrastructures in areas susceptible to floods. Consequently, implementing an Internet of Things (IoT)-based flood monitoring system becomes crucial. Existing flood monitoring systems lack a comprehensive and scalable IoT platform to collect real-time data from diverse sensors efficiently, visualize flood information, and provide accurate water level forecasts. This thesis proposes a complete system designed to address the challenges associated with efficient data collection and flood monitoring from diverse IoT sensors. Our proposition involves creating and deploying a centralized system known as HYDROSIGHT, which facilitates the real-time gathering, monitoring, and visualization of flooding-related sensor data. HYDROSIGHT system also provides a log monitoring feature for effective debugging and troubleshooting. The IoT environment for flood monitoring and prediction system was designed to promote sustainability and autonomy by preferring sensors with minimal footprints and compatibility with solar panels. The system architecture leverages a 4G network for seamless data transmission. To validate the practical applicability of the proposed design,HYDROSIGHT system was tested at two municipalities of Quebec, namely Terrebonne, and Lac-Supérieur. In addition, the platform was also deployed at the Ericsson facility in Montreal to test the 5G capabilities. The deployment in these locations allowed us to evaluate the performance and effectiveness of the HYDROSIGHT system in a real flood monitoring environment. In addition to implementing the IoT testbed, a preliminary machine learning tool was developed on water level forecasting. In this experiment, we opted for an online machine-learning approach, recognizing the significance of real-time updates and low computational resources of IoT devices. Leveraging the constantly updating data from HYDROSIGHT, we trained and tested our online machine-learning model, enhancing its forecasting capabilities. We conducted a comparative analysis to understand the advantages of online machine learning over traditional batch learning. This analysis involved examining the water level forecasting results obtained from both methods using time series data from the HYDROSIGHT system deployed at Lac-Supérieur in Quebec.

Les inondations sont une préoccupation majeure avec un potentiel de risques importants pour la sécurité publique ainsi qu'un impact économique et social négatif. Pour développer un modèle hydrodynamique permettant de cartographier et d'évaluer les risques d'inondation, un modèle d'élévation est un élément essentiel. La grande disponibilité de données de télédétection multisources facilite la création d'un modèle numérique d'élévation topo-bathymétrique (TBDEM). Cependant, il peut être très difficile de créer un modèle d'élévation à haute résolution homogène adapté à la cartographie des inondations en raison des divergences entre les données topographiques et bathymétriques causées par des changements temporels, des systèmes de référence horizontaux et verticaux différents, et des différences significatives en termes de résolution, incertitude et zone de couverture. Cette étude présente une méthodologie qui élargit les études précédentes axées sur la cartographie côtière à basse résolution en résolvant les différences spatiales et temporelles des jeux de données multisources tout en maintenant l'intégrité de la morphologie des berges et de l'environnement proche du rivage. Ceci est réalisé en appliquant une nouvelle méthodologie de fusion qui est mieux adaptée aux sources de données en jeu. Une méthode de moindre coût est appliquée aux données topographiques alors qu'une méthode de feathering est appliquée aux données bathymétriques. Pour ce qui est de la zone intermédiaire à l'interface de la terre et de l'eau, des transects sont utilisés pour interpoler entre les données manquantes afin de garantir l'intégrité du littoral. Enfin, une méthode de krigeage empirique bayésien appliqué à l'ensemble des données permet de produire une surface sans discontinuité accompagnée d'une surface d'erreur pour analyser l'incertitude en chaque point du modèle. Des données LiDAR aéroporté ainsi que des données de bathymétrie multifaisceau de la section supérieure du fleuve Saint-Laurent au Québec, Canada ont été combinées en utilisant la méthodologie proposée. Le TBDEM produit dans cette étude constitue une meilleure représentation que les modèles précédents et minimise l'erreur dans les données. La capacité de ce TBDEM à être plus performant que les modèles précédents dans les simulations hydrodynamiques sera testée dans des études futures en utilisant des événements de crue enregistrés précédemment.

The canalized southernmost section of Wonscotonach (the Don River) in Tkarón:to (Toronto), also known as The Narrows, is a highly disturbed urban natural landscape. Following the 1886 Don Improvement Project, the Keating Channel, and today the Port Lands Revitalization and Flood Protection Project, these Lands have been harnessed and developed through settler colonization to tame and control the once-winding river. This research-creation—in the form of a curated online exhibition and written thesis—presents a critical (re)reading of the notion of improvement, becoming allied to the pre-colonial landscape and the knowledge it carried. This exhibition and thesis develop the concept of the meander, inspired by the non-linear trajectory of the pre-canalized Don River, as a model for the curatorial. The curatorial process of improvement becomes a wall, and the river meanders still began before the global COVID-19 pandemic and, subsequently, was derailed in March 2020. The exhibition’s final form was unknowable throughout much of the curatorial process. Thus, following the meander as a research-creation technique, the curatorial process, exhibitionary structure, and content had to adapt through lingering uncertainty. This thesis, contributing to the theoretical and practical knowledge of research-creation, looks to intersections with the curatorial following the theoretical underpinnings of Erin Manning and Brian Massumi, Natalie Loveless and Stefanie Springgay and Sarah E. Truman. As a project untethered from institutional timelines and normative requirements to ‘know a project in advance,’ as well as the conventions of a physical exhibition, this research-creation manifested through process-led, creative and exploratory techniques (such as walking and drawing) and slowed pace allowed by the COVID-19 pandemic’s reframing of time. This research-creation exhibition and written thesis develop a responsive and resilient curatorial process deeply indebted to Land-based knowledge.

RÉSUMÉ: Les inondations sont considérées comme l'un des risques naturels les plus dangereux au monde. Plusieurs pays souffrent des conséquences néfastes des inondations. Au Canada, plusieurs provinces ont subi des inondations au cours du siècle dernier. Par exemple, la rivière des Outaouais a été confrontée à de nombreuses inondations comme en 2017 et 2019. La population d'Ottawa continue à augmenter d'une année à l'autre. C'est pour cela que nous avons choisi la rivière des Outaouais comme étude de cas pour ce projet dans le but de protéger la société contre les risques causés par les inondations. Les pays adoptent plusieurs solutions basées sur différentes méthodes afin de minimiser les dommages causés par les crues. La plupart des scientifiques s'accordent que la prévision des crues est la meilleure façon de limiter les conséquences des crues. Les systèmes de prévision des crues sont indispensables dans les pays fréquemment confrontés à des crues. Ils visent à fournir un long délai d'exécution et à fournir aux autorités et aux décideurs des informations suffisantes. Par conséquent, ils auront suffisamment de temps pour prendre les mesures adéquates pour sauver la vie de la population et limiter les catastrophes économiques dues aux inondations. ABSTRACT: Floods are one of the most catastrophic natural disasters in Canada and around the world that can cause loss of life and damages to properties and infrastructures. Saguenay flood (1996), southern Alberta flood (2013), and Ottawa floods (2017, 2019), are a few examples of Canadian floods with tremendous socio-economic impacts. Flood forecasting and predicting its characteristics (e.g., its magnitude and extent) has an important role in preventing and mitigating such flood impacts. Particularly, short-term forecasting is crucial for early warning systems and emergency response to floods. This study presents an integrated hydraulic-hydrologic modeling system for flood prediction. In this system, the Delft3D two-dimensional hydrodynamic model is connected with a HEC-HMS hydrologic model and observation data to provide an automatic exchange of data and results. Delft3D and HEC-HMS were chosen for this study because they were widely used and provided good results. In addition, they were applied in several flood forecasting studies. The prediction weather data and watershed characteristics provide input to the hydrological model to predict streamflow conditions, which are then automatically fed into the hydrodynamic model. The hydrodynamic model simulates the flood characteristics such as water level, 2D depth-averaged velocity field, and flood extent.

The frequency of natural hazards in North America presents a significant challenge for governments due to the damages they cause to the environment. Floods are severe hydrological events caused by spring snowmelt and intense rain events. Flood frequency analysis studies assumes that annual peak flood events occur independently of each other, regardless of previous flood events (the independent and identically distributed (i.i.d.) assumption); however, annual peak flood records do not necessarily appear to conform to these assumptions. First, a review of the literature on the effects of climate oscillations on extreme flood frequencies in North America was conducted. Then, the i.i.d. flood event assumption was tested by analyzing the effects of the Pacific Decadal Oscillation (PDO) and Atlantic Multidecadal Oscillation (AMO) on 250 naturally flowing annual peak flood records across the entire western North American margin. Using permutation tests on quantile-quantile (Q-Q) plots, I found that the PDO has a greater impact on the magnitude of annual peak floods than the AMO. Twenty-six percent of the gauges have higher magnitude annual floods depending on the PDO phase (p < 0.1). Next, I examined the interacting effects of the PDO and AMO on the frequencies of lower and upper quartile annual peak floods, and found reinforcing, cancelling, and dominating effects. Since these two climate oscillations have significant effects on the magnitudes of annual peak floods, the i.i.d. assumption does not hold. Hence, I advocate for the need to re-assess baseline flood analysis in western North America to improve flood management strategies.

Abstract: Accurate spring flow forecasts, primarily driven by snow accumulation and melt processes, are essential for decision-makers aiming to optimize hydro-electricity production and mitigate potential flood damages in snow-dominated regions. By integrating snowpack data from diverse sources (in-situ, remote sensing, and reanalysis) with modeled snow-related state variables through Data Assimilation (DA), there is the potential to leverage both modeling and observations for more accurate estimation of the resulting spring flow. However, challenges such as the lack of sufficient snow station networks and issues with optical and microwave sensors to provide Snow Water Equivalent (SWE) can hinder progress, particularly in regions with heterogeneous snowpack characteristics. Overcoming these challenges is crucial to realizing the full potential of snow DA in advancing spring flow forecasts. This thesis aims to optimize spring flow forecasting effectiveness using snow DA, where snow-related observations are incorporated into hydrological models. It investigates the best hydrological model structure (a lumped model: HSAMI, and a distributed model: HYDROTEL) for leveraging distributed SWE data provided by SNODAS (SNOw Data Assimilation System) dataset, which serves as the observation. This DA framework is applied on a large, heterogeneous northern Québec watershed (Outardes 4). It does so by updating SWE model states of HYDROTEL and HSAMI using the Ensemble Kalman Filter (EnKF) DA method. The simulated spring flow are then compared to observed spring flow data. Another scope of the thesis is to compare two reanalysis products with varying spatial resolutions and SWE representations (SNODAS and ERA5-Land) for improving 1-day spring flow forecasts in HYDROTEL over a smaller southern Québec watershed (Au Saumon). Finally, the thesis assesses the forecasting skill of the optimal model-observation combination over an extended 30-day forecast horizon using various probabilistic and deterministic metrics. The forecasting skills are evaluated in terms of SWE estimations during the snowpack accumulation and melt periods, and their impacts on spring flow. Among the two hydrological models considered (i.e., HYDROTEL and HSAMI), HYDROTEL proves to be a better candidate to unlock the full potential of distributed SNODAS SWE dataset through DA over a large watershed with spatially variable SWE. This is seen by improved 1-day spring flow forecasts metrics over many years (2014-2017). From the observation source point of view, SNODAS DA results in a more consistent 1-day spring flow forecasts compared to ERA5-Land over Au Saumon. However, the improvements in 1-day spring flow forecasts induced by SNODAS DA are comparatively modest over Au Saumon compared to Outardes 4, with NSE changing from 0.44 to 0.45, 0.31 to 0.34, and 0.59 to 0.61 for the 2014-2017 time period. This could be rooted in alignments between the physiographic characteristics of the watershed and the frequency of DA updates. The results obtained from the first two chapters of the thesis provide a snow DA framework with the capability to improve short-term and mid-term SWE forecasts with varying influence over the forecast horizon given the snowpack period considered (i.e., formation and ablation). The improved SWE estimations lead to increased accuracy and better uncertainty representations, as measured by Nash-Sutcliffe Efficiency (NSE), Relative Bias (RB), and Continuous Ranked Probability Score (CRPS), in spring flow forecasts for the Outardes 4 watershed over the study period (2014-2017).

While impressive results have been achieved in the well-known fields where Deep Learning allowed for breakthroughs such as computer vision, its impact on different older areas is still vastly unexplored. In Computational Fluid Dynamics and especially in Flood Modeling, many phenomena are very high-dimensional, and predictions require the use of numerical simulations, which can be, while very robust and tested, computationally heavy and may not prove useful in the context of real-time predictions. This issue led to various attempts at developing Reduced-Order Modeling techniques, both intrusive and non-intrusive. One recent relevant addition is a combination of Proper Orthogonal Decomposition with Deep Neural Networks (POD-NN). Yet, to our knowledge, little has been performed in implementing uncertainty-aware regression tools in the example of the POD-NN framework. In this work, we aim at comparing different novel methods addressing uncertainty quantification in Neural Networks, pushing forward the POD-NN concept with Deep Ensembles and Bayesian Neural Networks, which we first test on benchmark problems, and then apply to a real-life application: flooding predictions in the Mille-Iles river in Laval, QC, Canada. Building a non-intrusive surrogate model, able to know when it doesn’t know, is still an open research area as far as neural networks are concerned.

Extreme flood events continue to be one of the most threatening natural disasters around the world due to their pronounced social, environmental and economic impacts. Changes in the magnitude and frequency of floods have been documented during the last years, and it is expected that a changing climate will continue to affect their occurrence. Therefore, understanding the impacts of climate change through hydroclimatic simulations has become essential to prepare adaptation strategies for the future. However, the confidence in flood projections is still low due to the considerable uncertainties associated with their simulations, and the complexity of local features influencing these events. The main objective of this doctoral thesis is thus to improve our understanding of the modelling uncertainties associated with the generation of flood projections as well as evaluating strategies to reduce these uncertainties to increase our confidence in flood simulations. To address the main objective, this project aimed at (1) quantifying the uncertainty contributions of different elements involved in the modelling chain used to produce flood projections and, (2) evaluating the effects of different strategies to reduce the uncertainties associated with climate and hydrological models in regions with diverse hydroclimatic conditions. A total of 96 basins located in Quebec (basins dominated by snow-related processes) and Mexico (basins dominated by rain-related processes), covering a wide range of climatic and hydrological regimes were included in the study. The first stage consisted in decomposing the uncertainty contributions of four main uncertainty sources involved in the generation of flood projections: (1) climate models, (2) post-processing methods, (3) hydrological models, and (4) probability distributions used in flood frequency analyses. A variance decomposition method allowed quantifying and ranking the influence of each uncertainty source on floods over the two regions studied and by seasons. The results showed that the uncertainty contributions of each source vary over the different regions and seasons. Regions and seasons dominated by rain showed climate models as the main uncertainty source, while those dominated by snowmelt showed hydrological models as the main uncertainty contributor. These findings not only show the dangers of relying on single climate and hydrological models, but also underline the importance of regional uncertainty analyses. The second stage of this research project focused in evaluating strategies to reduce the uncertainties arising from hydrological models on flood projections. This stage includes two steps: (1) the analysis of the reliability of hydrological model’s calibration under a changing climate and (2) the evaluation of the effects of weighting hydrological simulations on flood projections. To address the first part, different calibration strategies were tested and evaluated using five conceptual lumped hydrological models under contrasting climate conditions with datasets lengths varying from 2 up to 21 years. The results revealed that the climatic conditions of the calibration data have larger impacts on hydrological model’s performance than the lengths of the climate time series. Moreover, changes on precipitation generally showed greater impacts than changes in temperature across all the different basins. These results suggest that shorter calibration and validation periods that are more representative of possible changes in climatic conditions could be more appropriate for climate change impact studies. Following these findings, the effects of different weighting strategies based on the robustness of hydrological models (in contrasting climatic conditions) were assessed on flood projections of the different studied basins. Weighting the five hydrological models based on their robustness showed some improvements over the traditional equal-weighting approach, particularly over warmer and drier conditions. Moreover, the results showed that the difference between these approaches was more pronounced over flood projections, as contrasting flood magnitudes and climate change signals were observed between both approaches. Additional analyses performed over four selected basins using a semi-distributed and more physically-based hydrological model suggested that this type of models might have an added value when simulating low-flows, and high flows on small basins (of about 500 km2). These results highlight once again the importance of working with ensembles of hydrological models and presents the potential impacts of weighting hydrological models on climate change impact studies. The final stage of this study focused on evaluating the impacts of weighting climate simulations on flood projections. The different weighting strategies tested showed that weighting climate simulations can improve the mean hydrograph representation compared to the traditional model “democracy” approach. This improvement was mainly observed with a weighting approach proposed in this thesis that evaluates the skill of the seasonal simulated streamflow against observations. The results also revealed that weighting climate simulations based on their performance can: (1) impact the floods magnitudes, (2) impact the climate change signals, and (3) reduce the uncertainty spreads of the resulting flood projection. These effects were particularly clear over rain-dominated basins, where climate modelling uncertainty plays a main role. These finding emphasize the need to reconsider the traditional climate model democracy approach, especially when studying processes with higher levels of climatic uncertainty. Finally, the implications of the obtained results were discussed. This section puts the main findings into perspective and identifies different ways forward to keep improving the understanding of climate change impacts in hydrology and increasing our confidence on flood projections that are essential to guide adaptation strategies for the future.