The Amazon river system has by far the largest streamflow of any river on Earth. Annual mean rainfall of over 2000 mm/yr is fed by atmospheric convergence of moisture from the tropical Atlantic, as well as recycling of evapotranspiration. A significant fraction of this precipitation runs off into river channels and is routed to the ocean. Amazon discharge into the Atlantic Ocean affects ocean salinity over a significant area, with consequent effects on sea surface temperature and ocean-atmosphere interactions. Despite this clear qualitative picture, quantitative understanding of the Amazon water cycle is limited, due to uncertainties in flux and storage terms of the water balance equation.

Here, we use satellite, reanalysis and in situ data of the atmosphere, land, and ocean to assess how accurately the Amazon water budget can be closed. This is achieved primarily through comparing three different estimates of Amazon streamflow based on: in situ measurements; atmospheric moisture convergence; precipitation and evaporation. We also discuss the reliability of GRACE water storage change data on different time scales. Finally, we use ocean salinity data as a novel test of the different streamflow estimates. This work is vital to better understand strengths and weaknesses of water cycle simulation in Earth system models.