Accidents and sabotages are kinds of environmental disasters that constitute a growing threat to marine ecosystems. Recent events disclosed to the public audience the importance of developing methodologies and systems that can accurately and continuously monitor such events. Sensor networks have been successfully applied to early warning and environmental monitoring both in terrestrial and marine contexts. In this paper we propose an architecture for continuous monitoring of marine disasters. After the detection of suspect substances in a sea area by means of its concentration in water, the first goal of this approach is to follow the contour of actual affected area by means of reconfigurable sensor networks and a centralized intelligent system. Such systems are able to estimate substance concentrations (functions that variate in space and time) by means of advanced geostatistical techniques, mobile sensor devices and evolutionary computing.