Using statistical learning to build better Earth System Models

点击翻译按钮获取中文摘要

Climate change caused by emissions of greenhouse gases presents an existential threat to society, industry and ecosystems around the world, and particularly in cold regions like Canada. Most Canadians will experience climate change at local scales, through changes to temperature, wind and rainfall patterns near their regions, cities, lakes and rivers. Climate scientists use sophisticated computer models to make projections of how global climate will respond to increasing greenhouse gas concentrations during the 21st century. However, future projections at the scale of individual Canadian river basins are highly uncertain, because models often disagree on whether future changes in precipitation and runoff will increase, or decrease, water availability. A major cause of the uncertainty is the grid box spacing of the models, known as the spatial resolution, which computational resources limit to about 100 km on each side. Decision-makers such as water managers need reliable river basin-scale projections with much finer grid spacing (around 10 km) to inform and adapt their management practices and infrastructure planning. Therefore, our inability as climate scientists to provide this information presents a major barrier to climate change adaptation in Canada, and beyond. The long-term goal of my research program is to improve the quality and efficiency of climate models to deliver global projections of climate change for Canada at a spatial resolution that is better suited to support decision-making activities. The first objective of the research is to develop and apply novel and efficient computing technologies, including methods based on artificial intelligence, to make it easier for climate scientists to produce climate projections that are useful for decision-makers. A second objective is to apply these high-resolution models to investigate the processes causing the uncertainty in future projections, such as snow accumulation and melt, or how clouds interact with pollution particles and sunlight. This ambitious research program represents a state-of-the-art fusion of modern earth system modelling and artificial intelligence methods, that has not been attempted before within a University environment in Canada. The research program will deliver essential training in climate science, modelling and artificial intelligence to a team of graduate and undergraduate students at the University of Waterloo. Graduates will exit the program with sophisticated and highly-marketable technical skills related to big data that are in high demand across Canada, as government agencies, NGOs and private industry undertake the next phase of evidence-based decision-making for climate change adaptation. The research and training outcomes will deliver new tools and technologies that will directly benefit government labs developing climate models, and all Canadians by improving our nation's capacity to develop resilient solutions to climate change at the local scale.