Date of Award

May 2024

Degree Type

Thesis

Degree Name

Master of Science

Department

Atmospheric Science

First Advisor

Clark Evans

Committee Members

Paul Roebber, Sergey Kravtsov

Abstract

Tropical cyclones (TC) typically decrease in intensity upon interacting with land because of increased surface roughness and decreased surface evaporation. However, several studies have documented cases in which TCs maintain their intensity or even intensify over land within non- or weakly baroclinic environments. Yet, our understanding of the precise physical processes that support maintenance or intensification over land in non- or weakly baroclinic environments remains limited, and the predictive skill for these outcomes has yet to be quantified.

We begin this process by quantifying the predictive skill and forecast uncertainty of the overland intensification of North Atlantic Tropical Storm Erin in 2007 using a 50-member ensemble of free forecasts initialized from the output of an ensemble adjustment Kalman Filter-based cycled data assimilation system using the Data Assimilation Research Testbed software and Advanced Research Weather Research and Forecasting model. The ensemble outputs are then analyzed using ensemble sensitivity analysis (to provide meaningful physical insight into the relevant forecast sensitivities, even in environments where non-linear processes are important), ensemble subsetting (e.g., strong versus weak TCs), and others, to assess the sensitivity in overland intensity to finite-amplitude atmospheric variability. Additionally, simpler measures such as intensity variability across the ensemble are utilized as part of the analysis, which we compare to both over-water intensification cases and idealized simulations of overland intensity change.We then take different surface and vertical level observation types plot them for both the prior and posterior analyses to compare the observation diagnostics at each analysis time. Optimal performance is achieved when the RMSE and spread are close in magnitude to each other, which could be indicative of well-tuned observation error statistics (Romine et al. 2013). Along with the evaluation of different observation diagnostics, a comprehensive analysis of ensemble outputs is conducted (Figs. 13, 14). Members are categorized into GOOD and BAD members, to help delineate which members best represent Erin’s intensity late on 18 August and the early hours of 19 August 2007 (Figs. 15, 16), which are determined with the help of MSLP, where the good members display at least one closed isobar for an extended period of time in the simulation, and bad members experience near-immediate dissipation.

Download

Share

COinS