Tropical-cyclone-induced coastal flooding results from the combined action of multiple interdependent processes, whose temporal phasing and magnitude govern the severity of inundation. Conventional statistical frameworks typically model these processes using static event maxima, neglecting their temporal evolution and implicitly forcing co-occurrence of component extremes, which can lead to biased or physically inconsistent design-level estimates. To address this limitation, this study applies the Multivariate Spatio-Temporal Maxima with Temporal Exposure (MSTM-TE) framework, which explicitly embeds temporal coherence in the simulation of compound extremes. The framework is applied to a long-term tropical cyclone dataset for Guadeloupe in the French Antilles, reconstructing time series of significant wave height, peak wave period, and sea surface height to generate physically consistent synthetic storm events. From these simulations, total water level (TWL) is computed as the sum of sea surface height and wave run-up, providing an integrated metric of coastal flooding hazard.</p> <p>Results show that MSTM-TE reduces bias and ensemble variance in TWL return-level estimation relative to conventional multivariate and location-specific approaches, highlighting the importance of preserving intra-storm temporal structure. The reconstructed storm evolutions reveal systematic temporal alignment between peaks in TWL and a compound wave-energy variable, indicating that wave-induced run-up dominates design-level flooding in the study domain. These findings establish MSTM-TE as a physics-informed statistical framework that bridges data efficiency, spatio-temporal realism, and methodological efficiency, offering a robust pathway for compound coastal flood risk assessment under limited observational data.