A key goal of edge computing is to achieve “distributed sensing” out of data continuously generated from a multitude of interconnected physical devices. The traditional approach is to gather information into sparse collector devices by relying on hop-by-hop accumulation, but issues of reactivity and fragility naturally arise in scenarios with high mobility. We propose novel algorithms for dynamic data summarisation across space, supporting high reactivity and resilience by specific techniques maximising the speed at which information propagates towards collectors. Such algorithms support idempotent and arithmetic aggregation operators and, under reasonable network assumptions, are proved to achieve optimal reactivity. We provide evaluation via simulation: first in multiple scenarios showing improvement over the state of art, and then by a case study in edge data mining, which conveys the practical impact in higher-level distributed sensing patterns.