Understanding how the climate will behave months in advance is important for managing risks in areas like farming, energy, water use, and disaster planning. However, many current models struggle to make reliable forecasts beyond a few weeks. This project introduces a new artificial intelligence (AI) model called DUNE2, designed to predict weekly temperature changes across the continental United States up to one year in advance.

DUNE2 uses recent climate data, such as temperature, wind, ocean conditions, and soil type, to learn patterns and make future predictions. It is trained using high-resolution data from the ERA5 reanalysis, which combines real-world measurements with advanced models. The model is built using a combination of deep learning techniques that can capture both large-scale patterns and local details.

By improving long-range weather forecasts, DUNE2 could help communities, businesses, and governments make better decisions about how to prepare for the weather months in advance.