A5008759934- Oceanographic and Atmospheric Processes
- Meteorological Phenomena and Simulations
- Climate variability and models
- Reservoir Engineering and Simulation Methods
- Atmospheric and Environmental Gas Dynamics
- Model Reduction and Neural Networks
- Ocean Waves and Remote Sensing
- Computational Physics and Python Applications
- Geophysics and Gravity Measurements
- Hydrological Forecasting Using AI
- Electric Vehicles and Infrastructure
- Geology and Paleoclimatology Research
- Gaussian Processes and Bayesian Inference
- Plant Water Relations and Carbon Dynamics
- Smart Grid Energy Management
- Climate change and permafrost
- Marine and coastal ecosystems
- Marine and fisheries research
- Millimeter-Wave Propagation and Modeling
- Cryospheric studies and observations
- Explainable Artificial Intelligence (XAI)
- Bayesian Modeling and Causal Inference
- Energy Load and Power Forecasting
- Geological and Geophysical Studies
- Microwave and Dielectric Measurement Techniques
Institut Pierre-Simon Laplace
2020-2025
Institut de Recherche pour le Développement
2018-2025
Sorbonne Université
2020-2023
Laboratoire d'Océanographie et du Climat : Expérimentations et Approches Numériques
2020-2022
Laboratoire des Sciences du Climat et de l'Environnement
2020-2022
CEA Paris-Saclay
2020-2021
Commissariat à l'Énergie Atomique et aux Énergies Alternatives
2020-2021
Université Paris Cité
2021
Université Grenoble Alpes
2018-2020
Institut des Géosciences de l'Environnement
2018-2020
In light of growing interest in data-driven methods for oceanic, atmospheric, and climate sciences, this work focuses on the field data assimilation presents analog (AnDA). The proposed framework produces a reconstruction system dynamics fully manner where no explicit knowledge dynamical model is required. Instead, representative catalog trajectories assumed to be available. Based catalog, combines nonparametric sampling using forecasting with ensemble-based techniques. This study explores...
This work presents EddyNet, a deep learning based architecture for automated eddy detection and classification from Sea Surface Height (SSH) maps provided by the Copernicus Marine Environment Monitoring Service (CMEMS). EddyNet consists of convolutional encoder-decoder followed pixel-wise layer. The output is map with same size input where pixels have following labels {`0': Non eddy, `1': anticyclonic `2': cyclonic eddy}. Keras Python code, training datasets weights files are open-source...
Abstract The use of machine learning to build subgrid parametrizations for climate models is receiving growing attention. State‐of‐the‐art strategies address the problem as a supervised task and optimize algorithms that predict fluxes based on information from coarse resolution models. In practice, training data are generated higher numerical simulations transformed in order mimic simulations. By essence, these meet so‐called priori criteria. But actual purpose parametrization obtain good...
A physics informed approach is applied to neural networks for subgrid-scale scalar flux modeling. We show that several invariances of the transport equation are not enforced by existing parametric models, which reduce their interpretability and question application. new architecture embedding these as hard soft constraints proposed. Through different flow configurations, we proposed increase both performances generalization capabilities model.
Abstract The trustworthiness of neural networks is often challenged because they lack the ability to express uncertainty and explain their skill. This can be problematic given increasing use in high stakes decision‐making such as climate change applications. We address both issues by successfully implementing a Bayesian Neural Network (BNN), where parameters are distributions rather than deterministic, applying novel implementations explainable AI (XAI) techniques. analysis from BNN provides...
Ocean eddies, as a ubiquitous phenomenon of the global ocean, are extremely important for ocean energy and material exchanges. Therefore, efficient eddy detection tracking crucial advancing our understanding dynamics. This work presents framework automatic by leveraging state-of-the-art machine learning algorithms. First, we propose new convolutional neural network model multieddies detection. is capable extracting accurate boundary information eddies fitting gap between semantic context sea...
Abstract The North Atlantic ocean is key to climate through its role in heat transport and storage. Climate models suggest that the circulation weakening but physical drivers of this change are poorly constrained. Here, root mechanisms revealed with explicitly transparent machine learning (ML) method Tracking global Heating Ocean Regimes (THOR). Addressing fundamental question existence dynamical coherent regions, THOR identifies these their link distinct currents such as formation regions...
Abstract. Geoscientific models are facing increasing challenges to exploit growing datasets coming from remote sensing. Universal differential equations (UDEs), aided by differentiable programming, provide a new scientific modelling paradigm enabling both complex functional inversions potentially discover physical laws and data assimilation heterogeneous sparse observations. We demonstrate an application of UDEs as proof concept learn the creep component ice flow, i.e. nonlinear diffusivity...
This paper reflects the discussions of early and mid-career researchers (EMCRs) during World Climate Research Programme Open Science Conference 2023 EMCRs Symposium, to advance climate knowledge for greater transformative power in society impact on policy-making. These focused three key priority challenges: how produce robust, usable, used information at local scale; address research gaps Global South; could support policy-making with information. We present here our perspective these major...
Ocean General Circulation Models require extensive computational resources to reach equilibrium states, while deep learning emulators, despite offering fast predictions, lack the physical interpretability and long-term stability necessary for climate scientists understand sensitivity (to greenhouse gas emissions) mechanisms of abrupt % variability such as tipping points. We propose take best from both worlds by leveraging generative models produce physically consistent oceanic states that...
Abstract. In Senegal, the West African monsoon (WAM) season is characterized by pronounced subseasonal to seasonal (S2S) rainfall fluctuations in response complex interactions between large-scale atmospheric and oceanic variability patterns mesoscale convective systems. Indeed, general circulation models (GCMs) used development of S2S forecasting systems often struggle represent mechanisms yielding WAM predictability. This study explores potential machine learning (ML) approaches improve...
Water resources in mountainous regions are affected by climate change, necessitating accurate hydrological simulations to provide plausible future scenarios for effective management. This study evaluates the added value of a high-resolution regional model (RCM) projecting water under scenarios. The RegIPSL Earth system model, was employed simulate European South-West (SWE3) domain at convection-permitting scale with horizontal resolution 3 km. Precipitation and temperature outputs were...
Besides regional climate models (RCMs), there exist two main approaches to tackle the insufficient resolution of global models: emulators and statistical downscaling. While both are similar in techniques they use (statistical machine learning, ML, methods), differ their objectives underlying assumptions. Emulators intended provide a cost-effective alternative RCMs by emulating downscaling functions. Alternatively, (SD) learn empirical (observed) relationships that link set key large-scale...
Ocean General Circulation Models (hereafter OGCM) are critical to the study of past and present climate, production future projections. Unfortunately, they require large amounts computations at simulation time. On other hand, deep learning emulators trained on reanalyses starting deliver accurate short-term predictions, using comparatively small computational resources, yet struggle long term not interpretable do no't factor in uncertainty physical parameters. As a result, cannot be used by...
Satellite-derived products are of key importance for the high-resolution monitoring ocean surface on a global scale. Due to sensitivity spaceborne sensors atmospheric conditions as well associated spatio-temporal sampling, remote sensing data may be subject high-missing rates. The interpolation these remains challenge deliver L4 gridded end-users. Whereas operational mostly rely model-driven approaches, especially optimal based Gaussian process priors, availability large-scale observation...
From the recent developments of data-driven methods as a means to better exploit large-scale observation, simulation and reanalysis datasets for solving inverse problems, this study addresses improvement reconstruction higher-resolution Sea Level Anomaly (SLA) fields using analog strategies. This is stated an data assimilation issue, where models rely on patch-based Empirical Orthogonal Functions (EOF)-based representations circumvent curse dimensionality. We implement Observation System...
Abstract The Southern Ocean closes the global overturning circulation and is key to regulation of carbon, heat, biological production, sea level. However, dynamics general upwelling pathways remain poorly understood. Here, a physics-informed unsupervised machine learning framework using principled constraints used. A unifying proposed invoking semi-circumpolar supergyre south Antarctic circumpolar current: massive series leaking sub-gyres spanning Weddell Ross seas that are connected...
The objective of this study is to evaluate the potential for History Matching (HM) tune a climate system with multi-scale dynamics. By considering toy model, namely, two-scale Lorenz96 model and producing experiments in perfect-model setting, we explore detail how several built-in choices need be carefully tested. We also demonstrate importance introducing physical expertise range parameters, priori running HM. Finally revisit classical procedure tuning, that consists tuning slow fast...
This paper presents an online reinforcement learning based application which increases the revenue of one particular electric vehicles (EV) station, connected to a renewable source energy. Moreover, proposed adapts changes in trends station's average number customers and their types. Most parameters model are simulated stochastically algorithm used is Q-learning algorithm. A computer simulation was implemented demonstrates confirms utility model.
The Southern Ocean closes the global overturning circulation and is key to regulation of carbon heat, biological production, sea level. However, dynamics general upwelling pathways remain poorly understood. Here, a unifying framework proposed invoking semi-circumpolar `supergyre' south Antarctic circumpolar current: massive series  ‘leaking’ sub-gyres spanning Weddell Ross seas that are connected maintained via rough topography acts as scaffolding. supergyre...
Identifying fronts manually from satellite images is a tedious and subjective task. Accordingly, edge detection algorithms are introduced for automatic of fronts. However, traditional cannot be applied to cloud-contaminated images, because missing data caused by occasional cloud coverage interferes with front detection. To diminish this risk, letter proposes new algorithm quick an accurate instant sea surface temperature (SST) image, instead depending on the daily or weekly averaged SST...
Progress within physical oceanography has been concurrent with the increasing sophistication of tools available for its study. The incorporation machine learning (ML) techniques offers exciting possibilities advancing capacity and speed established methods also making substantial serendipitous discoveries. Beyond vast amounts complex data ubiquitous in many modern scientific fields, study ocean poses a combination unique challenges that ML can help address. observational is largely spatially...
Marine hydrological elements are of vital importance in marine surveys. The evolution these can have a profound effect on the relationship between human activities and hydrology. Therefore, detection explanation laws urgently needed. In this paper, novel method, named Evolution Trend Recognition (ETR), is proposed to recognize trend ocean fronts, being most important information dynamic process. we focus task ocean-front classification. A classification algorithm first for recognizing trend,...
The ever increasing geophysical data streams pouring from earth observation satellite missions and numerical simulations along with the development of dedicated big infrastructure advocate for truly exploiting potential these datasets, through novel data-driven strategies, to deliver enhanced satellite-derived gapfilled products partial observations. We here demonstrate relevance analog assimilation (AnDA) an application reconstruction cloud-free level-4 gridded Sea Surface Temperature...