Gathering physical data in a marine environment has always been expensive. Sensored buoys are commonly deployed to gather a sampling of data, but the cost and number required for comprehensive coverage of a region rapidly becomes exorbitant. Remote sensing technology has made the wholesale acquisition of maritime physical readings both practical and economical. However, natural interference still causes widespread, unpredictable 'drop-outs' of readings throughout the remotely-sensed field. Further, periods of high interference and low data acquisition often coincide with severe weather conditions, when maritime accidents with significant economic or ecological effects are most likely – and when up-to-date surface readings are most needed. The reconstruction of missing readings by correlating past behavior and proxy ocean current readings to resulting surface currents can be explored through the use of artificial neural networks, models that have not traditionally been applied to spatially-distributed phenomena. Spatial interpolation of missing surface ocean current data collected in the Gulf of Maine is achieved by combining clustering techniques and backpropagation: separate neural models are trained on prior ocean current behavior to predict missing readings within each cluster-partitioned region. Neural model results within these behavioral regions demonstrate that this divide and conquer approach can reliably reconstruct missing surface current measurements within the field. Error reduction rates on the order of 70% over naïve model results, and positive performance comparisons to standard spatial approximation techniques support the validity of a neural data mining approach for spatial applications.