Project Details
Projekt
Volcanic Impact on surface climate (VolClim)
Applicants
Dr. Christopher Kadow; Dr. Claudia Timmreck
Subject Area
Atmospheric Science
Term
since 2019
Project identifier
Deutsche Forschungsgemeinschaft (DFG) - Project number 398006378
When the next large climate relevant volcano erupts, large changes in the Earth system will have an impact on seasonal and decadal climate predictability. It is therefore essential to ensure that we are prepared for future volcanic events under present day conditions but also under future warmer climate. An assessment of the climate impact of potential future volcanic eruptions is however only possible if the main processes and pathways through which volcanoes affect climate variability and predictability are fully understood. This is in particular crucial for processes that are controlled largely by dynamical changes (Northern Hemisphere Winter circulation, tropical hydroclimate). Artificial Intelligence/Machine Learning (AI/ML) has become a powerful tool for prediction in many domains. A natural question to ask is whether these data-driven methods could be used to predict the climate response to volcanic eruptions in advance. Here, we combine our previous and ongoing work on the volcanic impact on climate with our expertise with AI/ML techniques. In VolClim, we will develop a volAI system which consists on three neural networks based on deep learning applications: 1) predAI to predict the surface climate response to volcanic eruptions; 2) ensAI to enlarge single model run to ensemble size, and 3) resAI to downscale low resolution model runs with a super resolution approach and ultra high resolution simulations. predAI and resAI will be based on a convolutional neural network (CNN), ensAI on a generative adversarial network (GAN). As all approaches are data driven, several training periods are necessary to investigate the process and results to improve certain prediction capabilities of the volAI. Applying volAI, we will investigate if the seasonal to annual impact on tropical hydroclimate and NH surface winter climate of a potential future volcano is more predictable or precise, if AI/ML methods get trained using multi ESM data. We exploit this for volcanic eruptions of different eruption strengths and seasons, geographical latitudes and background conditions. We also investigate if AI/ML will help us to characterize the relative contribution of initial conditions and radiative volcanic forcing. In the future, in a warmer world the climate impact of natural forcing might change dramatically. We will therefore test if our volAI system trained to volcanic eruptions under present day conditions has the same skill for a future eruption that happened under global warming.
DFG Programme
Research Units
