Today, nearly half of the global population lives within 150 km of a coastline. As continued coastal development coincides with rising sea levels and more frequent and intense storms, the incidence, and cost of natural disasters is expected to rise. In the United States, recent studies have shown that current flood hazard estimates widely under-predict actual flood losses in coastal areas, resulting in billions of dollars of avoidable damage. However, historical data for tropical cyclones (especially storm surge and precipitation) is often limited or insufficient for analyzing or predicting tropical cyclone impacts. Furthermore, floodplains, which drive policy decisions regarding local planning, new development, flood insurance, and flood mitigation, often oversimplify the complex processes associated with flooding in the coastal zone. This dissertation aims to improve existing methods for predicting flood hazards and associated risk in highly urbanized coastal watersheds by integrating currently available models into a multi-hazard framework. The research is divided into three phases: (1) characterizing storm surge behavior in Galveston Bay, (2) establishing boundary conditions for floodplain modeling, and (3) flood hazard and risk analysis. In the first phase, the coupled Simulating WAves Nearshore and ADvanced CIRCulation (SWAN+ADCIRC) Model is used to simulate storm surge in Galveston Bay. The results demonstrate that storm surge in the bay is dominated by local wind direction and landfall location. Furthermore, counterclockwise rotating winds cause the highest storm surges to occur in the heavily populated evacuation zones on the north and west shores of Galveston Bay. Thus, subsequent research focused on the Clear Creek Watershed which encompasses much of the heavily urbanized west side of Galveston Bay. The second phase focuses on modeling probable combinations of surge and precipitation for Clear Creek, located on the west side of Galveston Bay. To do so, a Non-parametric Bayesian Network based on copulas is built and combined with a 1-D bay model to stochastically simulate a large number of synthetic storms in the Gulf of Mexico. The Bayesian Network is computationally inexpensive and takes into consideration five tropical cyclone characteristics at landfall: windspeed, angle of approach, distance to landfall, radius of maximum winds, and forward velocity. The resulting network is flexible and can be easily expanded to incorporate additional data as it becomes available. In the final phase of research, flood hazard and risk in the watershed are modeled using a distributed hydrologic modeling software, Vflo(R), in combination with the hydraulic model, HEC-RAS. The dissertation culminates with a longitudinal assessment of the evolution of flood risk since 1970 in an urbanizing coastal watershed. Utilizing the proposed framework, the impact of localized land use/land cover (LULC) change on the spatial extent of flooding in the watershed and the underlying flood hazard structure are quantified. The results demonstrate that increases in impervious cover substantially increase the spatial extent of the floodplain, as well as the depth and frequency of flooding in neighborhoods within the 1\% floodplain. Finally, the analysis provides evidence that by incorporating physics-based distributed hydrologic models into floodplain studies, floodplain maps can be easily updated to reflect the most recent LULC information available. The methods presented in this dissertation have important implications for the development of mitigation strategies in coastal areas, such as deterring future development in flood prone areas and directing flood mitigation efforts in already flood prone communities.