Leveraging Advancements in Ensemble Data Assimilation to Assess Tropical Cyclone Hazards

Recent developments in ensemble data assimilation (DA) for numerical weather prediction have facilitated the ability to evaluate the impact of using non-Gaussian DA on modeling various weather phenomena. This presentation focuses on evaluating characteristics of tropical cyclone (TC) hazards by using metrics that capture the associated distribution of surface winds among experiments that differ only in DA methodology. Results shown capitalize on experimental ensemble DA configurations within NOAA’s Hurricane Analysis and Forecast System (HAFS) that encompass the entirety of the Atlantic and Eastern Pacific TC basins, cycling over a large, fixed regional domain. In contrast to TC-oriented operational modeling systems where cycling DA is performed on regional storm-following nests initialized only after individual TCs have been identified, the presented configurations do not limit data to regions within high-resolution nests, allowing land-based observations to provide benefit to long-term TC forecasts.

An assessment of TC characteristics will be provided using forecasts initialized from the operational DA method (Ensemble Kalman Filter – 3D Variational Hybrid) alongside non-Gaussian ensemble DA methods such as the localized particle filter and its variants. These emerging DA techniques show promise for known NWP issues related to nonlinear model dynamics, as existing operational DA strategies assume Gaussian and linear error characteristics which inhibit the optimal use of observations for initialization. Results reveal benefits for TC prediction that have previously been omitted due to a limited use of high-resolution ensemble DA over extensive domains. A comparison of ensemble-based DA methods highlights the importance of reliable probabilistic predictions for interpreting impacts to TC-focused metrics, such as track, intensity, and surface wind field extent. Probabilistic quantities such as ensemble mean and spread of various TC metrics will be shown, indicating a consistent increase in ensemble spread across all metrics assessed when using non-Gaussian ensemble DA. While impacts to TC track and intensity will be discussed, results focus on TC surface wind structure, as the push towards more hazard-based metrics becomes favored.