This project aims to improve the applicability, reliability, and robustness of cutting-edge artificial intelligence (AI) technology in the understanding of the impacts of climate change in alpine regions. Specifically, it will improve the mapping of alpine forests, which we will apply to aerial imagery taken over Valais since the 1940s. The project combines the cross-disciplinary expertise of two research groups respectively at EPFL Valais and Idiap: environmental science and machine learning.Context. This project is embedded in the ongoing numerical transformation of society, as computational tools bring new challenges and opportunities. Meanwhile, Valais’ unique and fragile alpine ecosystems are heavily impacted by climate change and the biodiversity crisis. These, together with changes in land use resulting from the economic transformation of the canton, make environmental monitoring a necessity. For example, tree habitats are shifting upwards, impacting local animal and plant communities, and transforming the landscape Moreover, extreme events such as droughts periods and landslides are increasingly common and severe, and alpine forests are known for their protective function against these hazards. Understanding forest dynamics and their causal links with land interventions and climate change is critical to define nature management policies. Emerging AI technologies have incredible potential in helping reach these objectives but several of their shortcomings remain to be addressed. These include limitations in reliability to generalize over time and across geographical locations. This project will ensure that the adoption of AI for environmental monitoring delivers reliable and positive outcomes for our canton.Methods. This project will develop new AI methods to interpret aerial imagery collected over diverse geographical areas and time periods. These will be applied to aerial imagery of taken over Valais since the 1940s, leading to the creation of an interactive map of the evolution of forests to the present day. Several scientific challenges need to be addressed, as generalization to novel contexts (e.g. time periods, locations) poses a major problem in machine learning. We propose an approach that leverages recent advances made by the applicants’ research groups in their respective fields of study [14, 15, 31, 32]. Impact. The first expected result is a software tool for the automated monitoring of alpine forest dynamics.It will enable a better understanding of their drivers, from agricultural land abandonment to climate change. These questions are particularly relevant for our canton but also on a global scale given the essential role that forests play for the conservation of biodiversity, recreation, protection against natural hazards, and carbon sequestration. We will also create a web portal where citizens can discover the evolution of alpine forests of Valais over time. We aim to collaborate with the Service des forêts, de la nature et du paysage to evaluate how these mapping tools can be integrated into their monitoring plans. We also expect scientific contributions on core topics in machine learning (robustness, generalization) with broad potential impact on academic research and on the adoption of reliable AI in various domains.