A multidisciplinary approach that combines field measurements, artificial neural networks, water balance analyses and hydrodynamic modelling was developed to investigate the water budget and renewal capacity of semi-closed coastal systems. The method was applied to the Lesina Lagoon, a micro-tidal lagoon in the southern Adriatic Sea (Italy). Surface water flux between the lagoon and the sea was determined by neural network prediction and used as input in the analysis. Strong seasonal variations in the water budget equation were predicted. Fresh water inputs estimated by the water balance analysis were used as forcing by a calibrated finite element model to describe the water circulation and transport time scale of the lagoon’s surface waters. The model highlighted the spatial heterogeneity of the renewal behaviour of the system, with a strong east–west water renewal time gradient. Knowledge of spatial distribution of water renewal times is crucial for understanding the lagoon’s renewal capacity and explaining the high spatial variability of the biogeochemistry of the Lesina Lagoon.