A Variational U-Net for Weather Forecasting

TL;DR

This paper introduces a Variational U-Net model that combines probabilistic autoencoding with detailed image recovery to improve weather forecasting accuracy, demonstrating its versatility across different domains.

Contribution

The paper presents a novel Variational U-Net architecture that integrates variational autoencoders with U-Net for enhanced weather prediction capabilities.

Findings

Achieved third place in Weather4cast competition

Demonstrated the model's applicability to weather and traffic domains

Showed improved prediction accuracy over previous methods

Abstract

Not only can discovering patterns and insights from atmospheric data enable more accurate weather predictions, but it may also provide valuable information to help tackle climate change. Weather4cast is an open competition that aims to evaluate machine learning algorithms' capability to predict future atmospheric states. Here, we describe our third-place solution to Weather4cast. We present a novel Variational U-Net that combines a Variational Autoencoder's ability to consider the probabilistic nature of data with a U-Net's ability to recover fine-grained details. This solution is an evolution from our fourth-place solution to Traffic4cast 2020 with many commonalities, suggesting its applicability to vastly different domains, such as weather and traffic.