The global ocean's near surface can be partitioned into distinct provinces on the basis of regional primary productivity and oceanography [1]. This ecological geography provides a valuable framework for understanding spatial variability in ecosystem function but has relevance only partway into the epipelagic zone (the top 200 m). The mesopelagic (200-1,000 m) makes up approximately 20% of the global ocean volume, plays important roles in biogeochemical cycling [2], and holds potentially huge fish resources [3-5]. It is, however, hidden from satellite observation, and a lack of globally consistent data has prevented development of a global-scale understanding. Acoustic deep scattering layers (DSLs) are prominent features of the mesopelagic. These vertically narrow (tens to hundreds of m) but horizontally extensive (continuous for tens to thousands of km) layers comprise fish and zooplankton and are readily detectable using echosounders. We have compiled a database of DSL characteristics globally. We show that DSL depth and acoustic backscattering intensity (a measure of biomass) can be modeled accurately using just surface primary productivity, temperature, and wind stress. Spatial variability in these environmental factors leads to a natural partition of the mesopelagic into ten distinct classes. These classes demark a more complex biogeography than the latitudinally banded schemes proposed before [6, 7]. Knowledge of how environmental factors influence the mesopelagic enables future change to be explored: we predict that by 2100 there will be widespread homogenization of mesopelagic communities and that mesopelagic biomass could increase by approximately 17%. The biomass increase requires increased trophic efficiency, which could arise because of ocean warming and DSL shallowing.