A5065243427- Atmospheric and Environmental Gas Dynamics
- Climate variability and models
- Meteorological Phenomena and Simulations
- Remote Sensing in Agriculture
- Gaussian Processes and Bayesian Inference
- Domain Adaptation and Few-Shot Learning
- Air Quality Monitoring and Forecasting
- Neural Networks and Applications
- Simulation Techniques and Applications
- demographic modeling and climate adaptation
- Atmospheric aerosols and clouds
- Advanced Neural Network Applications
- Atmospheric chemistry and aerosols
- Bayesian Modeling and Causal Inference
- Machine Learning and Data Classification
- Remote Sensing and LiDAR Applications
- Advanced Image and Video Retrieval Techniques
- Advanced Database Systems and Queries
- Fault Detection and Control Systems
- Combustion and flame dynamics
- Infrared Target Detection Methodologies
- Advanced Image Fusion Techniques
- Multi-Criteria Decision Making
- Soil Geostatistics and Mapping
- Advanced Combustion Engine Technologies
University of Oxford
2020-2024
CentraleSupélec
2020
Abstract Many different emission pathways exist that are compatible with the Paris climate agreement, and many more possible miss target. While some of most complex Earth System Models have simulated a small selection Shared Socioeconomic Pathways, it is impractical to use these expensive models fully explore space possibilities. Such explorations therefore mostly rely on one‐dimensional impulse response models, or simple pattern scaling approaches approximate physical given scenario. Here...
Abstract Understanding the complex dynamics of climate patterns under different anthropogenic emissions scenarios is crucial for predicting future environmental conditions and formulating sustainable policies. Using Dynamic Mode Decomposition with control (DMDc), we analyze surface air temperature from simulations to elucidate effects various climate-forcing agents. This improves upon previous DMD-based methods by including forcing information as a variable. Our study identifies both common...
Abstract Emulators, or reduced complexity climate models, are surrogate Earth system models (ESMs) that produce projections of key quantities with minimal computational resources. Using time‐series modeling more advanced machine learning techniques, data‐driven emulators have emerged as a promising avenue research, producing spatially resolved responses visually indistinguishable from state‐of‐the‐art ESMs. Yet, their lack physical interpretability limits wider adoption. In this work, we...
Many different emission pathways exist that are compatible with the Paris climate agreement, and many more possible miss target. While some of most complex Earth System Models have simulated a small selection Shared Socioeconomic Pathways, it is impractical to use these expensive models fully explore space possibilities. Such explorations therefore mostly rely on one-dimensional impulse response models, or simple pattern scaling approaches approximate physical given scenario. Here we present...
Refining low-resolution (LR) spatial fields with high-resolution (HR) information, often known as statistical downscaling, is challenging the diversity of datasets prevents direct matching observations. Yet, when LR samples are modeled aggregate conditional means HR respect to a mediating variable that globally observed, recovery underlying fine-grained field can be framed taking an "inverse" expectation, namely deconditioning problem. In this work, we propose Bayesian formulation which...
Emulators, or reduced complexity climate models, are surrogate Earth system models that produce projections of key quantities with minimal computational resources. Using time-series modelling more advanced machine learning techniques, data-driven emulators have emerged as a promising avenue research, producing spatially resolved responses visually indistinguishable from state-of-the-art models. Yet, their lack physical interpretability limits wider adoption. In this work, we introduce...
Mixup - a neural network regularization technique based on linear interpolation of labeled sample pairs has stood out by its capacity to improve model's robustness and generalizability through surprisingly simple formalism. However, extension the field object detection remains unclear as bounding boxes cannot be naively defined. In this paper, we propose leverage inherent region mapping structure anchors introduce mixup-driven training for proposal detectors. The proposed method is...
Abstract Aerosol-cloud interactions constitute the largest source of uncertainty in assessments anthropogenic climate change. This arises part from difficulty measuring vertical distributions aerosols, and only sporadic vertically resolved observations are available. We often have to settle for less informative aggregated proxies such as aerosol optical depth (AOD). In this work, we develop a framework disaggregation AOD into extinction profiles, that is, measure light throughout an...
We develop the framework of Indirect Query Bayesian Optimization (IQBO), a new class optimization problems where integrated feedback is given via conditional expectation unknown function $f$ to be optimized. The underlying distribution can and learned from data. goal find global optimum by adaptively querying observing in space transformed distribution. This motivated real-world applications one cannot access direct due privacy, hardware or computational constraints. propose Conditional...
Analyzing climate scenarios is crucial for quantifying uncertainties, identifying trends, and validating models. Objective statistical methods provide decision support policymakers, optimize resource allocation, enhance our understanding of complex dynamics. These tools offer a systematic quantitative framework effective decision-making policy formulation amid change, including accurate projections extreme events—a fundamental requirement Earth system modeling actionable future...
Out-of-distribution (OOD) generalisation is challenging because it involves not only learning from empirical data, but also deciding among various notions of generalisation, e.g., optimising the average-case risk, worst-case or interpolations thereof. While this choice should in principle be made by model operator like medical doctors, information might always available at training time. The institutional separation between machine learners and operators leads to arbitrary commitments...
Public satellite missions are commonly bound to a trade-off between spatial and temporal resolution as no single sensor provides fine-grained acquisitions with frequent coverage. This hinders their potential assist vegetation monitoring or humanitarian actions, which require detecting rapid detailed terrestrial surface changes. In this work, we probe the of deep generative models produce high-resolution optical imagery by fusing products different characteristics. We introduce dataset...
The recent explosion in applications of machine learning to satellite imagery often rely on visible images and therefore suffer from a lack data during the night. gap can be filled by employing available infra-red observations generate images. This work presents how deep applied successfully create those using U-Net based architectures. proposed methods show promising results, achieving structural similarity index (SSIM) up 86\% an independent test set providing visually convincing output...
A directed acyclic graph (DAG) provides valuable prior knowledge that is often discarded in regression tasks machine learning. We show the independences arising from presence of collider structures DAGs provide meaningful inductive biases, which constrain hypothesis space and improve predictive performance. introduce regression, a framework to incorporate probabilistic causal problem. When reproducing kernel Hilbert space, we prove strictly positive generalisation benefit under mild...
Emulators, or reduced complexity climate models, are surrogate Earth system models that produce projections of key quantities with minimal computational resources. Using time-series modeling more advanced machine learning techniques, statistically-driven emulators have emerged as a promising venue, producing spatially resolved responses visually indistinguishable from state-of-the-art models. Yet, their lack physical interpretability limits wider adoption. In fact, the exploration future...
Emulators, or reduced complexity climate models, are surrogate Earth system models that produce projections of key quantities with minimal computational resources. Using time-series modelling more advanced machine learning techniques, data-driven emulators have emerged as a promising avenue research, producing spatially resolved responses visually indistinguishable from state-of-the-art models. Yet, their lack physical interpretability limits wider adoption. In this work, we introduce...
<p>Exploration of future emissions scenarios mostly relies on one-dimensional impulse response models, or simple pattern scaling approaches to approximate the physical climate a given scenario. Such are unable reliably predict variables which respond non-linearly forcing (such as precipitation) and must rely heavily simplified representations e.g., aerosol, neglecting important spatial dependencies.</p><p>Here we present ClimateBench - benchmark...
Aerosol-cloud interactions constitute the largest source of uncertainty in assessments anthropogenic climate change. This arises part from difficulty measuring vertical distributions aerosols, and only sporadic vertically resolved observations are available. We often have to settle for less informative aggregated proxies such as aerosol optical depth (AOD). In this work, we develop a framework disaggregation AOD into extinction profiles, i.e. measure light throughout an atmospheric column,...
Aerosol-cloud interactions constitute the largest source of uncertainty in assessments anthropogenic climate change. This arises part from inability to observe aerosol amounts at cloud formation levels, and, more broadly, vertical distribution aerosols. Hence, we often have settle for less informative two-dimensional proxies, i.e. vertically aggregated data. In this work, formulate problem disaggregation profiles We propose some initial solutions such an aggregate output regression and...
Earth and Space Science Open Archive This preprint has been submitted to is under consideration at Journal of Advances in Modeling Systems (JAMES). ESSOAr a venue for early communication or feedback before peer review. Data may be preliminary.Learn more about preprints preprintOpen AccessYou are viewing an older version [v1]Go new versionClimateBench: A benchmark dataset data-driven climate projectionsAuthorsDuncanWatson-ParrisiDYuhanRaoDirkOliviéØyvindSelandPeer...
Aerosol-cloud interactions constitute the largest source of uncertainty in assessments anthropogenic climate change. This arises part from inability to observe aerosol amounts at cloud formation levels, and, more broadly, vertical distribution aerosols. Hence, we often have settle for less informative two-dimensional proxies, i.e. vertically aggregated data. In this work, formulate problem disaggregation profiles We propose some initial solutions such an aggregate output regression and...