In coastal catchments, co-occurring rainfall and storm surge can compound flood depth and extent through combined hydrodynamic effects, while remote co-occurrence of the two drivers can compound impacts, e.g., by straining emergency response resources. This study characterizes the spatial structure of rainfall–surge dependence in the Mullica River Basin, New Jersey, using a high-resolution gridded rainfall dataset and a high-resolution coastal water level reanalysis. Dependence between rainfall and the non-tidal residual (NTR) is quantified separately for tropical cyclone (TC) and non-TC events using Kendall's τ and fitted copulas. TC events exhibit stronger and more spatially variable dependence than non-TC events, and results are largely insensitive to coastal point choice except where sheltered from open-coast by a barrier island. Representing coastal forcing with total water level rather than NTR substantially weakened apparent TC dependence, underscoring the need to isolate the NTR in compound flood analyses. Joint return periods for the 10-year rainfall and the 10-year NTR design event are 30–60 years across most of the watershed, shorter than the 100-year period expected under independence and are governed primarily by copula tail dependence rather than correlation strength alone. Joint return-period estimates based on a locally composited rainfall record, constructed from mean hourly rainfall across AORC grid points within a ~30 km radius of the coastal point, more closely reproduce the basin-wide estimates than those based on either the rain gauge nearest to the coastal point or basin-averaged rainfall. A response-based framework for generating synthetic compound flood events using basin-averaged rainfall showed some systematic correlation bias, particularly for TC events, and underestimated short-return-period non-TC water levels driven by spring tides.