Statistical downscaling and climate change in the coastal zone

• Main Author: Obakrim, Said
• Other Authors: Institut de Recherche Mathématique de Rennes (IRMAR), Université de Rennes (UR)-Institut National des Sciences Appliquées - Rennes (INSA Rennes), Institut National des Sciences Appliquées (INSA)-Institut National des Sciences Appliquées (INSA)-École normale supérieure - Rennes (ENS Rennes)-Université de Rennes 2 (UR2)-Centre National de la Recherche Scientifique (CNRS)-Institut Agro Rennes Angers, Institut national d'enseignement supérieur pour l'agriculture, l'alimentation et l'environnement (Institut Agro)-Institut national d'enseignement supérieur pour l'agriculture, l'alimentation et l'environnement (Institut Agro), Université de Rennes, Valérie Monbet, Pierre Ailliot
• Format: Doctoral or Postdoctoral Thesis
• Language: English
• Published: HAL CCSD 2022
• Subjects: Downscaling; Sea state; Generalized Ridge; Mixture of experts; EM algorithm; Deep learning; Descente d'échelle; Etat de mer; Ridge généralisée; Mélange d'experts; Algorithme EM; Apprentissage profond; [INFO.INFO-DS]Computer Science [cs]/Data Structures and Algorithms [cs.DS]; [INFO.INFO-LG]Computer Science [cs]/Machine Learning [cs.LG]; North Atlantic
• Online Access: https://theses.hal.science/tel-03952800; https://theses.hal.science/tel-03952800/document; https://theses.hal.science/tel-03952800/file/OBAKRIM_Said.pdf

Ocean wave climate has a significant impact on human activities, and its understanding is socioeconomically and environmentally important. In this thesis, we are interested in characterizing sea state parameters such as significant wave height (Hs) using statistical and deep learning methods. In particular, we are interested in modeling the relationship between North Atlantic wind conditions and sea state parameters at a location in the Bay of Biscay. Given the multidimensionality of the wind data and the time-lagged relationship between wind conditions and waves, we first propose a general framework to select the relevant covariates that influence the significant wave height. After the preprocessing step, a regression model based on weather types is proposed to model the relationship between wind and waves. The weather types are constructed using a clustering algorithm, and then, for each weather type, a Ridge regression is fitted between the wind conditions and the significant wave height. The model predicts Hs well; however, it has some limitations, namely: (i) Ridge regression does not take into account that the covariates have a spatial structure; and (ii) the weather types are constructed a priori using a clustering algorithm, and they are not evaluated based on the prediction of Hs. Therefore, we propose an expectation-maximization (EM) algorithm to estimate the parameters of the generalized Ridge regression with spatial covariates. Then, to account for (i) and (ii), we propose a mixture of generalized Ridge experts estimated using a variational EM algorithm. This model is used as a weather-types-based regression model, and its performance is better than that of the original model.Finally, the last part of this thesis is devoted to developing deep learning methods for sea state parameters prediction. Le climat des vagues océaniques a un impact significatif sur les activités humaines, et sa compréhension est importante sur le plan socio-économique et environnemental. Dans cette thèse, nous nous ...