Enhancing Tropical Cyclone Ensemble Forecast Skill via the Collective Effect of Initial and Model Perturbations: The C-NFSVs Method

The skill of forecasting Tropical Cyclone (TC) Rapid Intensification (RI) is limited by inherent uncertainties in initial conditions and model physics. To address this, the C-NFSVs method integrates initial and model perturbations, accounting for their collective effects through the nonlinear forcing singular vector (NFSV; also known as CNOP-F) approach. In this study, we applied C-NFSVs to the Weather Research and Forecasting (WRF) model for TC ensemble forecasting across three resolutions, comparing it against O-NFSVs, which has proven superior to traditional stochastic physics schemes. Results reveal a significant resolution dependence, with the superiority of C-NFSVs maximizing at the convection-permitting scale. At this resolution, the C-NFSVs ensemble outperforms O-NFSVs for both deterministic and probabilistic metrics, and demonstrates significantly improved reliability. Notably, for the challenging prediction of RI events, C-NFSVs exhibits high discriminative skill, achieving an Area Under the ROC Curve (ROCA) of 0.80. A detailed examination of TC Hato attributes this success to capturing the evolution of the critical physical error chain, which progresses from thermodynamic priming and convective organization to the structural and dynamic response. Mechanistically, the results highlight the complementary roles of the two components: the initial component of C-NFSVs dominates the uncertainty of the dynamic structure in the early forecast stage, while the model component plays a primary role in maintaining the thermodynamic uncertainty of moisture and temperature fields throughout the forecast. This study validates the effectiveness and physical rationality of C-NFSVs in high-resolution ensembles, offering a promising strategy for enhancing the predictability of extreme weather events at convection-permitting scales.