In this paper we propose a re-thinking of ICT infrastructure to include a framework that exploits commodity many-core systems to evaluate models. The framework permits comparison, evaluation and improvement of competing and complementary models. Our proposal focuses on the computationally intensive tasks associated with near-field parallelism to process environmental conditions in real-time to deliver informed warnings to populations at risk. By keeping a human-in-the-loop, we include new services that support crowd-sourcing within the framework, allowing integration of sensor data with media-rich voluntary participant input. Monte Carlo simulations of relevant ocean models highlight necessary precursors and likelihoods of potential threats. This paper provides a survey of existing systems, both from tsunami modeling and other domains with similar real-time constraints, and evaluates the applicability of our proposed framework, PREDICT, for near-field tsunami early warning.