L'énergie éolienne est aujourd'hui la source d'énergie renouvelable en plus forte expansion. Le caractère variable et partiellement contrôlable de sa production complexifie la gestion du système électrique. L'utilisation dans divers processus de décision, de prédictions du niveau de production à des horizons de 2-3 jours, permet une meilleure intégration de cette ressource. Certaines situations donnent néanmoins lieu à des performances de prédiction insatisfaisantes. Des erreurs dans la prédiction de l'instant d'apparition de variations brusques et importantes de la production, peuvent être responsables d'importants déséquilibres énergétiques, et avoir un impact négatif sur la gestion du système électrique. L'objectif de cette thèse est de proposer des approches permettant d'une part de caractériser ces variations, et d'autre part de prédire et d'estimer l'incertitude dans l'instant de leur apparition. Dans un premier temps, nous étudions différentes formes de caractérisation de ces variations. Nous proposons un modèle de rupture permettant de représenter le caractère aléatoire dans la proximité des ruptures d'un signal, tout en tenant compte des aspects borné et non-stationnaire du processus de production. A partir de simulations issues de ce modèle, nous réalisons une étude paramétrique destinée à évaluer et comparer les performances de différents filtres et approches multi-échelles de détection. Dans un deuxième temps, nous proposons une approche de prédiction probabiliste de l'instant d'apparition d'une rupture, reposant sur l'utilisation de prévisions météorologiques ensemblistes. Leur conversion en puissance fournit différents scénarii de la production, à partir desquels sont agrégées les prédictions de l'instant d'apparition d'une rupture. L'incertitude associée est représentée à l'aide d'intervalles de confiance temporels et de probabilités estimées conditionnellement. Nous évaluons la fiabilité et la finesse de ces estimations sur la base de mesures de production provenant de différentes fermes éoliennes. / Today, wind energy is the fastest growing renewable energy source. The variable and partially controllable nature of wind power production causes difficulties in the management of power systems. Forecasts of wind power production 2-3 days ahead can facilitate its integration. Though, particular situations result in unsatisfactory prediction accuracy. Errors in forecasting the timing of large and sharp variations of wind power can result in large energy imbalances, with a negative impact on the management of a power system. The objective of this thesis is to propose approaches to characterize such variations, to forecast their timing, and to estimate the associated uncertainty. First, we study different alternatives in the characterization of wind power variations. We propose an edge model to represent the random nature of edge occurrence, along with representing appropriately the bounded and non-stationary aspects of the wind power production process. From simulations, we make a parametric study to evaluate and compare the performances of different filters and multi-scale edge detection approaches. Then, we propose a probabilistic forecasting approach of edge occurrence and timing, based on numerical weather prediction ensembles. Their conversion into power provides an ensemble of wind power scenarios from which the different forecast timings of an edge are combined. The associated uncertainty is represented through temporal confidence intervals with conditionally estimated probabilities of occurrence. We evaluate the reliability and resolution of those estimations based on power measurements from various real world case studies.

Deep artificial neural networks is a type of machine learning which can be used to find and utilize patterns in data. One of their many applications is as method for regression analysis. In this thesis deep artificial neural networks were implemented in the application of estimating the error of surface temperature forecasts as produced by a numerical weather prediction model. An ability to estimate the error of forecasts is synonymous with the ability to reduce forecast errors as the estimated error can be offset from the actual forecast. Six years of forecast data from the period 2010--2015 produced by the European Centre for Medium-Range Weather Forecasts' (ECMWF) numerical weather prediction model together with data from fourteen meteorological observational stations were used to train and evaluate error-predicting deep neural networks. The neural networks were able to reduce the forecast errors for all the locations that were tested to a varying extent. The largest reduction in error was by 83.0\% of the original error or a 16.7\degcs decrease in the mean-square error. The performance of the neural networks' error reduction ability was compared with that of a contemporary Kalman filter as implemented by the Swedish Meteorological and Hydrological Institute (SMHI). It was shown that the neural network implementation had superior performance for six out of seven of the evaluated stations where the Kalman filter had marginally better performance at one station.

Les modèles de prévision opérationnelle du temps résolvent numériquement les équations de la mécanique des fluides en calculant l'évolution de champs (pression, température, humidité, vitesses) définis comme moyennes horizontales à l'échelle des mailles d'une grille (et à différents niveaux verticaux).<p><p>Les processus d'échelle inférieure à la maille jouent néanmoins un rôle essentiel dans les transferts et les bilans de chaleur, humidité et quantité de mouvement. Les paramétrisations physiques visent à évaluer les termes de source correspondant à ces phénomènes, et apparaissant dans les équations des champs moyens aux points de grille.<p><p>Lorsque l'on diminue la taille des mailles afin de représenter plus finement l'évolution des phénomènes atmosphériques, certaines hypothèses utilisées dans ces paramétrisations perdent leur validité. Le problème se pose surtout quand la taille des mailles passe en dessous d'une dizaine de kilomètres, se rapprochant de la taille des grands systèmes de nuages convectifs (systèmes orageux, lignes de grain).<p><p>Ce travail s'inscrit dans le cadre des développements du modèle à mailles fines ARPÈGE ALADIN, utilisé par une douzaine de pays pour l'élaboration de prévisions à courte échéance (jusque 48 heures).<p><p>Nous décrivons d'abord l'ensemble des paramétrisations physiques du modèle.<p>Suit une analyse détaillée de la paramétrisation actuelle de la convection profonde. Nous présentons également notre contribution personnelle à celle ci, concernant l'entraînement de la quantité de mouvement horizontale dans le nuage convectif.<p>Nous faisons ressortir les principaux points faibles ou hypothèses nécessitant des mailles de grandes dimensions, et dégageons les voies pour de nouveaux développements.<p>Nous approfondissons ensuite deux des aspects sortis de cette discussion: l'usage de variables pronostiques de l'activité convective, et la prise en compte de différences entre l'environnement immédiat du nuage et les valeurs des champs à grande échelle. Ceci nous conduit à la réalisation et la mise en œuvre d'un schéma pronostique de la convection profonde.<p>A ce schéma devraient encore s'ajouter une paramétrisation pronostique des phases condensées suspendues (actuellement en cours de développement par d'autres personnes) et quelques autres améliorations que nous proposons.<p>Des tests de validation et de comportement du schéma pronostique ont été effectués en modèle à aire limitée à différentes résolutions et en modèle global. Dans ce dernier cas l'effet du nouveau schéma sur les bilans globaux est également examiné.<p>Ces expériences apportent un éclairage supplémentaire sur le comportement du schéma convectif et les problèmes de partage entre la schéma de convection profonde et le schéma de précipitation de grande échelle.<p><p>La présente étude fait donc le point sur le statut actuel des différentes paramétrisations du modèle, et propose des solutions pratiques pour améliorer la qualité de la représentation des phénomènes convectifs.<p><p>L'utilisation de mailles plus petites que 5 km nécessite enfin de lever l'hypothèse hydrostatique dans les équations de grande échelle, et nous esquissons les raffinements supplémentaires de la paramétrisation possibles dans ce cas.<p><p> / Doctorat en sciences appliquées / info:eu-repo/semantics/nonPublished

This diploma thesis introduces three different methods of creating a neural network binary classifier for the purpose of automated weather prediction with attribute pre-selection using association rules and correlation patters mining by the LISp-Miner system. First part of the thesis consists of collection of theoretical knowledge enabling the creation of such predictive model, whereas the second part describes the creation of the model itself using the CRISP-DM methodology. Final part of the thesis analyses the performance of created classifiers and concludes the proposed methods and their possible benefits over training the network without attribute pre-selection.

Die hohe raumzeitliche Variabilität von konvektiven Wolken hat erhebliche Auswirkungen auf die Quantifizierung des Wolkenstrahlungseffektes. Da konvektive Wolken in atmosphärischen Modellen üblicherweise parametrisiert werden müssen, sind Beobachtungsdaten notwendig, um deren Variabilität sowie Modellunsicherheiten zu quantifizieren. Das Ziel der vorliegenden Dissertation ist die Charakterisierung der raumzeitlichen Variabilität von warmen konvektiven Wolkenfeldern mithilfe von Meteosat Beobachtungen sowie deren Anwendbarkeit für die Modellevaluierung. Verschiedene Metriken wurden untersucht, um Unsicherheiten in Modell- und Satellitendaten sowie ihre Limitierungen zu quantifizieren. Mithilfe des hochaufgelösten sichtbaren (HRV) Kanals von Meteosat wurde eine Wolkenmaske entwickelt, welche mit 1×2 km² die Auflösung der operationellen Wolkenmaske von 3×6 km² deutlich übertrifft. Diese ermöglicht eine verbesserte Charakterisierung von kleinskaligen Wolken und bietet eine wichtige Grundlage für die Weiterentwicklung von satellitengestützten Wolkenalgorithmen. Für die Untersuchung der Lebenszyklen konvektiver Wolkenfelder wurde ein Tracking-Algorithmus entwickelt. Die raumzeitliche Entwicklung des Flüssigwasserpfads (LWP) wurde sowohl in einer Eulerschen Betrachtungsweise als auch entlang Lagrange’scher Trajektorien analysiert. Für die Wolkenfelder ergab sich eine charakteristische Längenskala von 7 km. Als Maß für die Wolkenlebenszeit ergab sich eine Lagrange’sche Dekorrelationszeit von 31 min. Unter Berücksichtigung des HRV Kanals verringern sich die Dekorrelationsskalen signifikant, was auf eine Sensitivität gegenüber der räumlichen Auflösung hindeutet. Für eine Quantifizierung dieser Sensitivität wurden Simulationen des ICON-LEM Modells mit einer Auflösung von bis zu 156 m berücksichtigt. Verbunden mit einem zwei- bis vierfach geringeren konvektiven Bedeckungsgrad besitzen die simulierten Wolken bei dieser hohen Auflösung deutlich größere LWP Werte. Diese Unterschiede verschwinden im Wesentlichen, wenn die simulierten Wolkenfelder auf die optische Auflösung von Meteosat gemittelt werden. Die Verteilungen der Wolkengrößen zeigen einen deutlichen Abfall für Größen unterhalb der 8- bis 10-fachen Modellauflösung, was der effektive Auflösung des Modells entspricht. Dies impliziert, dass eine noch höhere Auflösung wünschenswert wäre, damit mit ICON-LEM Wolkenprozesse unterhalb der 1 km-Skala realistisch simuliert werden können. Diese Skala wird zukünftig erfreulicherweise vom Meteosat der dritten Generation abgedeckt. Dies wird ein entscheidender Schritt für ein verbessertes Verständnis von kleinskaligen Wolkeneffekten sowie für die Parametrisierung von Konvektion in NWP und Klimamodellen sein.

This diploma thesis deals with the comparison of the results of simulations of the numerical model WRF in the prediction mode for 9 schemes of boundary layer parameterization and in the climatic mode for 4 selected schemes. The first part of the work is devoted to the WRF model and especially its options for model physics with a focus on boundary layer schemes. The second part describes the experimental setup of the performed simulations. The third part then compares the obtained results for the prediction and climate mode with the measured data.

MODEL OF ERROR COVARIANCES FOR THE ASSIMILATION OF RADAR REFLECTIVITY INTO NWP MODEL Predicting events with a severe convection is not easy due to the small spatial scale and rapid development of this phenomenon. But being able to predict such events is important in view of the dangerous phenomena that accompany these events, such as flash floods, strong winds, hailstorms or atmospheric electricity. Improved forecast can be achieved by more precisely defined initial conditions that enter the model. These data must match the scale of the studied phenomenon. Therefore, radar data is used in this case. Although the NWP model should describe real processes due to the simplifications and approximations the model's behavior does not entirely correspond the reality. Therefore, if we want the model to generate precipitation, we must ensure that the values of the model variables and their relationship are such that the process is started. To find out these relationships, we want to use a covariant model. In this paper, we focused on the correlation analysis of the model variables in the regions of convection between radar reflection, its conversion to the intensity of precipitation and other model variables. The COSMO data with a horizontal resolution of 2.8 km were used, which were describing approximately...

Built-up areas represent an artifiial impait to natural environment with large spatial variability and speiifi meihaniit radiationt thermal and ihemiial properties. Despite of inireasing horizontal resolution of numeriial weather prediition modelst the impait of loial built-up area on mesosynoptiv weather phenomena is still not well resolved. Therefore it is neiessary to use some of urban environment modelst whiih were designed to parameterize speiifi urban prosiessest not expliiitly resolved inside the grid box. In the thesis main urban iharaiteristiis are explained (impait on the struiture of boundary layert radiation and heat balanie of urban environment or urban heat island)t basii priniiples of urbanization appliiation in the numeriial weather model are desiribedt as well as different urban parameterizations available in numeriial model WRFe (Weather Reseaih and Feoreiasting). Number of validation experiments were performed for summer and winter episode in non-hydrostatii mode at 3t3 km resolutiont where different urban parametrizationst antropogenii heat adjustment and impait of mosaii land-use were tested. April 2018 Prague weather foreiast was verifiated in ionsideration of urban heat island.

The use of numerical weather prediction (NWP) has brought significant improvements to severe weather outbreak forecasting; however, determination of the primary mode of severe weather (in particular tornadic and nontornadic outbreaks) continues to be a challenge. Uncertainty in model runs contributes to forecasting difficulty; therefore it is beneficial to a forecaster to understand the sources and magnitude of uncertainty in a severe weather forecast. This research examines the impact of data assimilation, microphysics parameterizations, and planetary boundary layer (PBL) physics parameterizations on severe weather forecast accuracy and model variability, both at a mesoscale and synoptic-scale level. NWP model simulations of twenty United States tornadic and twenty nontornadic outbreaks are generated. In the first research phase, each case is modeled with three different modes of data assimilation and a control. In the second phase, each event is modeled with 15 combinations of physics parameterizations: five microphysics and three PBL, all of which were designed to perform well in convective weather situations. A learning machine technique known as a support vector machine (SVM) is used to predict outbreak mode for each run for both the data assimilated model simulations and the different parameterization simulations. Parameters determined to be significant for outbreak discrimination are extracted from the model simulations and input to the SVM, which issues a diagnosis of outbreak type (tornadic or nontornadic) for each model run. In the third phase, standard synoptic parameters are extracted from the model simulations and a k-means cluster analysis is performed on tornadic and nontornadic outbreak data sets to generate synoptically distinct clusters representing atmospheric conditions found in each type of outbreak. Variations among the synoptic features in each cluster are examined across the varied physics parameterization and data assimilation runs. Phase I found that conventional and HIRS-4 radiance assimilation performs best of all examined assimilation variations by lowering false alarm ratios relative to other runs. Phase II found that the selection of PBL physics produces greater spread in the SVM classification ability. Phase III found that data assimilation generates greater model changes in the strength of synoptic-scale features than either microphysics or PBL physics parameterization.

The primary goal of this research is to evaluate and identify proper calibration approaches, skillful hydrological models, and suitable weather forecast inputs to improve the accuracy and reliability of hydrological forecasting in different types of watersheds. The research started by formulating an approach that examined single- and multi-site, and single- and multi-objective optimization methods for calibrating an event-based hydrological model to improve flood prediction in a semi-urban catchment. Then it assessed whether reservoir inflow in a large complex watershed could be accurately and reliably forecasted by simple lumped, medium-level distributed, or advanced land-surface based hydrological models. Then it is followed by a comparison of multiple combinations of hydrological models and weather forecast inputs to identify the best possible model-input integration for an enhanced short-range flood forecasting in a semi-urban catchment. In the end, Numerical Weather Predictions (NWPs) with different spatial and temporal resolutions were evaluated across Canada’s varied geographical environments to find candidate precipitation input products for improved flood forecasting. Results indicated that aggregating the objective functions across multiple sites into a single objective function provided better representative parameter sets of a semi-distributed hydrological model for an enhanced peak flow simulation. Proficient lumped hydrological models with proper forecast inputs appeared to show better hydrological forecast performance than distributed and land-surface models in two distinct watersheds. For example, forcing the simple lumped model (SACSMA) with bias-corrected ensemble inputs offered a reliable reservoir inflow forecast in a sizeable complex Prairie watershed; and a combination of the lumped model (MACHBV) with the high-resolution weather forecast input (HRDPS) provided skillful and economically viable short-term flood forecasts in a small semi-urban catchment. The comprehensive verification has identified low-resolution NWPs (GEFSv2 and GFS) over Western and Central parts of Canada and high-resolution NWPs (HRRR and HRDPS) in Southern Ontario regions that have a promising potential for forecasting the timing, intensity, and volume of floods. / Thesis / Doctor of Philosophy (PhD) / Accurate hydrological models and inputs play essential roles in creating a successful flood forecasting and early warning system. The main objective of this research is to identify adequately calibrated hydrological models and skillful weather forecast inputs to improve the accuracy of hydrological forecasting in various watershed landscapes. The key contributions include: (1) A finding that a combination of efficient optimization tools with a series of calibration steps is essential in obtaining representative parameters sets of hydrological models; (2) Simple lumped hydrological models, if used appropriately, can provide accurate and reliable hydrological forecasts in different watershed types, besides being computationally efficient; and (3) Candidate weather forecast products identified in Canada’s diverse geographical regions can be used as inputs to hydrological models for improved flood forecasting. The findings from this thesis are expected to benefit hydrological forecasting centers and researchers working on model and input improvements.