Beta diversity---the variation among community compositions in a region---is a fundamental indicator of biodiversity. Despite a diverse set of measures to quantify beta diversity, most measures have posited that beta diversity is maximized when each community has one distinct species. However, this postulate has ignored the importance of non-additivity of ecological systems (i.e., a community with two species is ecologically different from two communities with one species). Here, to account for this, we provide a geometric approach to measure beta diversity as the hypervolume of the geometric embedding of a metacommunity. We show that the hypervolume measure is closely linked to and naturally extends previous information- and variation-based measures. In addition, our hypervolume approach provides a unified geometric framework for widely adopted extensions on the basic measure of beta diversity: the contribution of duplications in presence/absence data, temporal changes, turnover-nestedness decomposition, species similarity and functional complementarity, and community/species-specific contributions. We apply our new geometric measures to empirical data and address two long-standing questions on beta diversity (latitudinal pattern and sampling efforts) and present novel ecological insights. In sum, our geometric approach reconceptualizes beta diversity, synthesizes previous measures and is immediately applicable to existing data.