Machine Learning-Driven Climate Model Improvement and Uncertainty Quantification

EasyChair Preprint 14133

Abstract

Accurate climate modeling and uncertainty quantification are crucial for understanding future climate scenarios and informing policy decisions. This research explores the integration of machine learning techniques to enhance the performance of climate models and improve the quantification of uncertainties. We employ advanced machine learning algorithms, such as deep learning and ensemble methods, to refine parameterizations, identify patterns, and correct biases in existing climate models. By leveraging large datasets from historical climate observations, satellite data, and climate simulations, we develop machine learning-driven models that can capture complex climate dynamics with higher fidelity. Additionally, we focus on improving uncertainty quantification through probabilistic models and techniques like Bayesian neural networks and Gaussian processes. These methods provide a more robust estimation of prediction uncertainties, offering valuable insights into the confidence levels of different climate projections. The study demonstrates significant improvements in model accuracy and uncertainty quantification, paving the way for more reliable climate predictions. The findings underscore the potential of machine learning to transform climate science, contributing to better-informed climate adaptation and mitigation strategies.

Keyphrases: Bayesian Neural Networks, Climate dynamics, Gaussian processes, adaptation strategies, bias correction, climate modeling, climate predictions, deep learning, ensemble methods, machine learning, parameterization, probabilistic models, uncertainty quantification

@booklet{EasyChair:14133,
  author    = {Joseph Oluwaseyi and Dylan Stilinki and Kaledio Potter},
  title     = {Machine Learning-Driven Climate Model Improvement and Uncertainty Quantification},
  howpublished = {EasyChair Preprint 14133},
  year      = {EasyChair, 2024}}