In the past decades, coastal compound floods (CCF) caused significant losses to coastal communities. To develop an accurate and complete probabilistic framework for assessing coastal hazards, the U.S. Army Corps of Engineers established the Coastal Hazard System’s Probabilistic Coastal Hazard Analysis (PCHA) framework. The current PCHA framework has focused primarily on a subset of CCF drivers, particularly storm surges.This dissertation documents four studies that contribute to efforts to advance and extend the PCHA for CCF analysis. In the first study, machine learning-based data imputation models are developed to fill in missing records in the historical storm dataset. The performance of the machine learning-based data imputation models under different parameterizations are assessed considering statistical and physical factors and also compared against existing prediction models. In the second study, a series of Joint Probability Method (JPM) assumptions to model the dependence among tropical cyclone (TC) atmospheric parameters are comparatively investigated. JPM assumptions considered in the analysis include parameter independence, partial dependence, and full dependence. Candidate full-dependence models include meta-Gaussian copula and vine copulas combining linear-circular copulas. Emphases are put on modeling the circular behavior of storm heading and its dependencies with other linear parameters. Full dependence models are compared based on the predicted probability of large CPD and RMW combinations. In the third study, a Bayesian Network (BN) of multiple coastal hazards is constructed, where the conditional probability tables (CPT) of TC atmospheric parameters are computed using copula. A deaggregation methodology is developed to identify the dominant TC for significant coastal hazard events to support risk-informing and decision-making processes and refined analyses. In the fourth study, leveraging the aforementioned study results, the extended PCHA is leveraged to develop a multi-tiered BN CCF analysis framework. In this framework, multiple tiers of BN models with different complexities are designed for study cases with varying levels of resource availability. A case study is conducted in New Orleans, LA and a series of joint distribution, numerical, machine learning, and experimental models are used to compute CPTs needed for BNs.