Deepfaking ecosystem response to climate extremes
Accurately predicting how ecosystems will respond to future climate extremes has been a challenge due to the limitations of current vegetation models and the lack of precise data to constrain them. To overcome these obstacles, the DeepV project proposes the development of data-driven, spatially explicit ecosystem response models using remote sensing data and conditional Generative Adversarial Networks (cGANs). By harnessing the power of cGANs, DeepV aims to generate realistic satellite image time series of ecosystem response based on environmental conditions and hydro-meteorological data, enabling climate-to-ecosystem response rule establishment and ecosystem sensitivity assessment.
Remote Sensing-Driven Downscaling Solutions for Antarctica
Understanding and projecting changes in key Antarctic climate processes—such as surface melt and surface mass balance (SMB)—remains difficult due to the limited spatial detail of current climate models and sparse observational data. ClimaVision addresses this challenge by developing a novel, physically informed downscaling framework that combines remote sensing data with cutting-edge super-resolution deep learning techniques. By embedding physical constraints into machine learning models and leveraging diverse satellite datasets, the project will produce high-resolution climate outputs that better capture local variability across Antarctica. These improved datasets will support more reliable assessments of ice sheet behavior and its role in future sea-level change.