The Climate Change AI Virtual Summer School 2026 is a five-week online program that looks at how artificial intelligence can help address climate change. Running from July 19 to August 22, 2026, the program features lectures from experts and hands-on tutorials. Topics include the basics of AI and climate change, responsible and ethical AI, and real-world applications in areas like energy, transportation, biodiversity, urban planning, public health, weather, and disaster risk management. The summer school is open to people from around the world, including researchers, practitioners, decision-makers, and students from academia, industry, civil society, and the public sector. Participants can join live or recorded lectures and work through tutorials at their own pace. Registration is now open.
CCAI Tutorial: PiggyCast – Improving Weather Prediction Accuracy through a Stacking-Based Ensemble AI Approach
Speakers: Jusiah Kimani, Oliver Angelil, Chris Toumping, Steffen Knoblauch
Accurate weather prediction is vital for effective disaster readiness, resource management, and societal resilience. Traditionally, weather and climate forecasts have depended on dynamical, physics-based Numerical Weather Prediction (NWP) models, which explicitly model weather processes by solving the governing equations of fluid dynamics and thermodynamics. While these models have achieved remarkable skill, their computational complexity and sensitivity to initial conditions present challenges, especially for high-resolution and long-range forecasts. Recent advances in statistical and machine learning (ML) techniques have positioned them as powerful complements to traditional NWP models. These methods, often described under the broader umbrella of Artificial Intelligence Weather Prediction (AIWP) models, can efficiently learn complex relationships from large datasets and correct systematic biases in dynamical model outputs while remaining computationally efficient in generating forecasts.To harness the strengths of both paradigms, hybrid forecasting systems have gained popularity. Hybrid models deliberately integrate concepts from dynamical (physics-based) and data-driven (artificial intelligence or statistical) models, aiming to enhance forecast skill across a range of hydroclimatic and meteorological variables and events, such as temperature, rainfall, streamflow, and extreme weather.
This study builds on these developments, recommending an ensemble machine learning model trained on top of the forecasts of NWP, AIWP and hybrid models to demonstrate that such a combined approach can surpass the predictive accuracy of any single base model. The ensemble model is termed PiggyCast (a portmanteau of the words piggyback and forecast).
Piggyback here means „to use something that already exists or has already been done successfully to do something else quickly or effectively“.
By the end of this tutorial, the users will:
- Develop and assess an ensemble machine learning model through stacking forecasts from numerical, AI-based and hybrid weather prediction models for enhanced predictive accuracy.
- Investigate the effect of input features on the trained ensemble model for interpretability and explainability of the forecasting process.
Schedule: Lectures will be presented live, as well as recorded. Tutorials will be fully asynchronous.
Language: The Virtual Summer School will be conducted in English, with live automatic subtitles available in various languages via the Zoom platform.
Registration and program information: www.climatechange.ai/events/summer_school