One of the important applications of global climate models is to predict anticipated changes in the statistical properties of extreme events. Tropical cyclones are among the extreme events with the greatest socioeconomic impacts in the United States and other regions of the world. Landfalling hurricanes cause significant loss of property and life along the Atlantic and Pacific coasts of North America. Although coarse-resolution global climate models are incapable of simulating individual hurricanes accurately, they do exhibit significant skill in simulating the interannual and decadal variations in the aggregate statistics of hurricanes in the Atlantic basin, when provided the observed sea surface temperature as the boundary condition. We propose to analyze and validate simulated tropical cyclone activity in the Energy Exascale Earth System Model (E3SM), with a focus on tropical cyclones in the Northern Hemisphere.

As global climate models approach horizontal spatial resolutions of 25km, their ability to simulate the statistical properties of tropical cyclones becomes an important validation metric. The U.S. CLIVAR Hurricane Working Group recently carried out an intercomparison of tropical cyclone characteristics as simulated by climate models and found that models are indeed able to reproduce the gross features of the geographical distribution of observed global tropical cyclone frequency. However, most models are not able to reproduce the detailed spatial structure of tropical cyclone tracks over the North Atlantic and other regions. In general, regionally-aggregated measures of tropical cyclone activity turn out to be much more predictable than local tropical cyclone occurrences. 

One of the challenges in simulating the spatial distribution of tropical cyclone track density in global climate models is the effect of climate bias. The genesis and evolution of tropical cyclones is quite sensitive to the large-scale background flow. For example, excessive vertical wind shear can inhibit the development of tropical cyclones. Since atmospheric flow biases can develop within a few weeks from the start of a simulation, it becomes difficult to distinguish between the flow bias effect and other possible deficiencies in the climate model, such as errors in subgrid parameterizations or poor spatial resolution. To address this problem, we propose to use an initialized ensembles approach, where a series of 14-day hindcasts is carried out using the atmospheric component of E3SM. The integrations will be initialized from atmospheric reanalyses every 3 days over the decadal period 2000-2009. By construction, the background flow in these hindcasts will be close to observations. Comparing the statistics of tropical cyclone simulations in the initialized ensemble to that in the control runs will allow us to isolate the impact of mean flow biases. Errors in the representation of fine-scale orographic features in certain areas, such as the Central American Gap Wind region, can also lead to biases in the simulation of tropical cyclones. We propose to use a mesh-refinement approach to better represent orographic features in this region and study its impact on tropical cyclone activity.