A5080734461- Climate variability and models
- Meteorological Phenomena and Simulations
- Energy Load and Power Forecasting
- Solar Radiation and Photovoltaics
- Machine Learning and Data Classification
- Domain Adaptation and Few-Shot Learning
- Adversarial Robustness in Machine Learning
- Data Visualization and Analytics
- Species Distribution and Climate Change
- Hydrological Forecasting Using AI
- Atmospheric and Environmental Gas Dynamics
- Climate Change Policy and Economics
- Urban Heat Island Mitigation
- Machine Learning and Algorithms
- Time Series Analysis and Forecasting
- Climate Change Communication and Perception
- Stochastic processes and financial applications
- Data Stream Mining Techniques
- Cryospheric studies and observations
University of California, San Diego
2023
While diffusion models can successfully generate data and make predictions, they are predominantly designed for static images. We propose an approach efficiently training probabilistic spatiotemporal forecasting, where generating stable accurate rollout forecasts remains challenging, Our method, DYffusion, leverages the temporal dynamics in data, directly coupling it with steps model. train a stochastic, time-conditioned interpolator forecaster network that mimic forward reverse processes of...
Aggregating multiple sources of weak supervision (WS) can ease the data-labeling bottleneck prevalent in many machine learning applications, by replacing tedious manual collection ground truth labels. Current state art approaches that do not use any labeled training data, however, require two separate modeling steps: Learning a probabilistic latent variable model based on WS -- making assumptions rarely hold practice followed downstream training. Importantly, first step does consider...
Deep learning-based models have recently outperformed state-of-the-art seasonal forecasting models, such as for predicting El Niño-Southern Oscillation (ENSO). However, current deep learning are based on convolutional neural networks which difficult to interpret and can fail model large-scale atmospheric patterns called teleconnections. Hence, we propose the application of spatiotemporal Graph Neural Networks (GNN) forecast ENSO at long lead times, finer granularity improved predictive skill...
Numerical simulations of Earth's weather and climate require substantial amounts computation. This has led to a growing interest in replacing subroutines that explicitly compute physical processes with approximate machine learning (ML) methods are fast at inference time. Within models, atmospheric radiative transfer (RT) calculations especially expensive. made them popular target for neural network-based emulators. However, prior work is hard compare due the lack comprehensive dataset...
Deep learning-based models have recently outperformed state-of-the-art seasonal forecasting models, such as for predicting El Niño-Southern Oscillation (ENSO). However, current deep learning are based on convolutional neural networks which difficult to interpret and can fail model large-scale atmospheric patterns. In comparison, graph (GNNs) capable of modeling spatial dependencies more interpretable due the explicit information flow through edge connections. We propose first application...
Data-driven deep learning models are on the verge of transforming global weather forecasting. It is an open question if this success can extend to climate modeling, where long inference rollouts and data complexity pose significant challenges. Here, we present first conditional generative model able produce ensemble simulations that accurate physically consistent. Our runs at 6-hourly time steps shown be stable for 10-year-long simulations. approach beats relevant baselines nearly reaches a...
In this study, we propose a solution to estimating system responses external forcings or perturbations. We utilize the Fluctuation-Dissipation Theorem (FDT) from statistical physics extract knowledge using an AI model that can rapidly produce scenarios for different by leveraging FDT and analyzing large dataset Earth System Models. Our model, AiBEDO, accurately captures complex effects of radiation perturbations on global regional surface climate, enabling faster exploration impacts...
Data programming (DP) has proven to be an attractive alternative costly hand-labeling of data. In DP, users encode domain knowledge into \emph{labeling functions} (LF), heuristics that label a subset the data noisily and may have complex dependencies. A model is then fit LFs produce estimate unknown class label. The effects misspecification on test set performance downstream classifier are understudied. This presents serious awareness gap practitioners, in particular since dependency...
The availability of training data remains a significant obstacle for the implementation machine learning in scientific applications. In particular, estimating how system might respond to external forcings or perturbations requires specialized labeled targeted simulations, which may be computationally intensive generate at scale. this study, we propose novel solution challenge by utilizing principle from statistical physics known as Fluctuation-Dissipation Theorem (FDT) discover knowledge...
Clouds have a significant impact on the Earth's climate system. They play vital role in modulating radiation budget and driving regional changes temperature precipitation. This makes clouds ideal for intervention techniques like Marine Cloud Brightening (MCB) which refers to modification cloud reflectivity, thereby cooling surrounding region. However, avoid unintended effects of MCB, we need better understanding complex response function. Designing testing such interventions scenarios with...
Clouds have a significant impact on the Earth's climate system. They play vital role in modulating radiation budget and driving regional changes temperature precipitation. This makes clouds ideal for intervention techniques like Marine Cloud Brightening (MCB) which refers to modification cloud reflectivity, thereby cooling surrounding region. However, avoid unintended effects of MCB, we need better understanding complex response function. Designing testing such interventions scenarios with...
<p>Deep learning-based models have been recently shown to be competitive with, or even outperform, state-of-the-art long range forecasting models, such as for projecting the El Niño-Southern Oscillation (ENSO). However, current deep learning are based on convolutional neural networks which difficult interpret and can fail model large-scale dependencies, teleconnections, that particularly important projections. Hence, we propose explicitly dependencies with Graph...