We consider the problem of efficiently processing subscription queries over data streams in large-scale interconnected sensor networks. We propose a scalable algorithm for distributed data stream processing, applicable on top of any platform granting access to interconnected sensor networks. We make use of a probabilistic algorithm to check whether subscriptions are subsumed by other subscriptions and thus can be pruned for more efficient processing. Our proposed methods are query driven, hence do not replicate data streams, but intelligently place join operators inside the global network of sources. We show by a performance evaluation using real world sensor data the suitability of our approach.