Wednesday, April 6, 2022 at 1:00pm CT
This video is currently private. To request the recorded talk please contact firstname.lastname@example.org.
Host: Patrick Heimbach
Title: Explore Tropical Cyclone Future Projections Using High-Resolution Numerical Simulations and Machine Learning Technique
Abstract: Tropical cyclones (TCs) pose great risks to individuals and societies, particularly in terms of their local impacts, underscoring the importance of a better understanding of their future change and predictability sources. In this talk, I will use three different modeling approaches, including the TC-permitting high-resolution global climate model, cloud-resolving regional coupled model and convolutional neural network (CNN) data-driven model, to explore TC future projections across various spatial and temporal scales.
1) There is no consensus on how climate change can influence global mean TC numbers and their tracks over individual ocean basins. Using multi-model ensembles of high-resolution atmosphere-only and fully coupled global climate model (GCM) experiments from the Coupled Model Intercomparison Project 6th phase (CMIP6) High Resolution Model Intercomparison Project (HighResMIP), we found that the global mean TC frequency generally decreases in the warming climate, primarily due to more stable atmosphere and increased vertical wind shear. However, the atmosphere-only and coupled simulations exhibit distinct spatial patterns of TC genesis and track projections. A signal-to-noise-maximizing empirical orthogonal function analysis reveals that climate internal multidecadal variability dominates TC projections in atmosphere-only simulations even over century timescales. In contrast, the anthropogenic external forcing determines future TC changes in the coupled simulations. Moreover, we showed that the increasing trend of extreme TC rainfall in the atmosphere-only simulations, are significantly underestimated compared to that in the coupled simulations.
2) It has been widely recognized that TC genesis and development require favorable large-scale environmental conditions. Based on these physical linkages, we trained ensembles of CNN to establish the nonlinear empirical relationship between seasonal TC activities and large-scale environments. Our data-driven model can accurately reproduce the historical TC observations, and yields even higher seasonal TC prediction skills compared to those predicted by operational centers. By applying this data-driven model to 20th century reanalysis data, we found a significant decreasing trend of seasonal TC activity in the western North Pacific since the 1830s, but moderate-weak increasing trend in the North Atlantic. Consistent with the TC projections directly resolved by the HighResMIP models, the application of CNN model to standard resolution CMIP6 simulations also suggests a decrease of global mean TC activity in the future climate. We also applied this data-driven model to decadal prediction and paleoclimate data to study TC variabilities.
3) To better evaluate the potential impacts of climate change on TC-induced damage to the human community, we employed the newly developed Regional Community Earth System Model (R-CESM) to make the hindcast simulations of those historically occurred TCs in cloud-resolving horizontal resolution. Using the “pseudo global warming” approach, we also performed R-CESM “future projection” experiments for Hurricane Harvey (2017). We found that Hurricane Harvey can induce 45% (31%) more rainfall if it were to occur in the 2090s (2050s). Moreover, Hurricane Harvey caused flooding inundation area in the Houston Clear Creek watershed is projected to increase by 32% (28%).