Project No: 26305720

Title: Quantifying and Understanding the Response of Extreme Convective Rainfall to Global Warming

Prof. Xiaoming SHI


Precipitation extremes, which can trigger landslides and floods, are expected to intensify under global warming and pose challenges to sustainable social development. However, reliable estimation of future tropical and subtropical precipitation extremes is beyond the capability of current climate models. This research project aims to use a machine learning-aided approach to selectively simulate extreme rainstorms in a warming climate and understand the expected changes with physical theories. Predictions of tropical and subtropical rainstorms using current climate models entail considerable uncertainty. Some models predict a 1% increase in the intensity of tropical precipitation extremes in response to 1° surface warming, whereas others suggest an increase of approximately 30%. Using observations to constrain model results can effectively reduce the range of uncertainty, but further improvement is still necessary. A direct cause of this uncertainty is the coarse resolution, which is typically 100 km horizontally, of climate models. Meanwhile, resolving tropical and subtropical rainstorms requires a resolution of approximately 1 km. Long-term global climate simulation at a kilometer-scale resolution remains impossible for today’s supercomputers. In this study, we will use support vector machines and cloud-resolving models (CRM) to enable high-resolution simulation of extreme rainfall events at climatic timescales. Current climate models with coarse grids misrepresent convective extreme rainfall events in simulations. We will use support vector machines to classify time slices from climate model simulations and identify large-scale circulation patterns that are likely to trigger severe rainfall events. Then, we will downscale the selected large-scale patterns with CRM to achieve reliable simulations of potential extreme precipitation events. We will produce an ensemble of extreme event simulations to address the uncertainty due to natural variability. Different climate models will be used to perform parent climate simulations. Various microphysics schemes, which may influence convection development, will be included for CRM simulations. The resulting ensemble of extreme rain events will allow us to quantify the uncertainty of our estimation and identify robust underlying trends. We will investigate the sensitivity of rainfall extremes to warming by partitioning precipitation change into thermodynamic and dynamic components; the latter refers to characteristics such as the intensity of vertical motions, storm structure, and occurrence frequency. Furthermore, we will evaluate the sensitivity of storm dynamics to warming from the perspectives of energetics, large-scale forcing, and moist dynamics. This study will achieve not only a robust estimation of severe convective precipitation under global warming but also improvement on the understanding of related changes.